engineering research, development and technology engineering

EngineeringResearch,Developmentand Technology


UCRL 53868-98UCRL 53868-98 FY 98FY 98

Eng

ineerin

gR

esearch,

Develo

pm

ent

and

Techn

olo

gy

•FY

98•

UC

RL

53868-98

Cover_qk 9/2/99 3:12 PM

DisclaimerThis document was prepared as an account of work sponsored by an agency of theUnited States Government. Neither the United States Government nor the University ofCalifornia nor any of their employees, makes any warranty, express or implied, orassumes any legal liability or responsibility for the accuracy, completeness, or usefulness of any information, apparatus, product, or process disclosed, or representsthat its use would not infringe privately owned rights. Reference herein to any specific commercial products, process, or service by trade name, trademark, manufacturer, orotherwise does not necessarily constitute or imply its endorsement, recommendation,or favoring by the United States Government or the University of California. The viewsand opinions of authors expressed herein do not necessarily state or reflect those of theUnited States Government or the University of California, and shall not be used for advertising or product endorsement purposes.

This report has been reproduceddirectly from the best available copy.

Available to DOE and DOE contractors from theOffice of Scientific and Technical Information

P.O. Box 62, Oak Ridge, TN 37831Prices available from (615) 576-8401, FTS 626-8401

Available to the public from theNational Technical Information Service

U.S. Department of Commerce5285 Port Royal Rd.,Springfield, VA 22161

Manuscript Date February 1999Distribution Category UC-706



UCRL 53868-98UCRL 53868-98 FY 98FY 98

FY98 title page 9/1/99 2:54 PM Page 1

Introduction

Spiros Dimolitsas, Associate Director for Engineering

1. Center for Complex Distributed Systems

OverviewRobert J. Deri, Center Leader

Accelerating the Development of Petaflop Applications and SystemsAnthony J. De Groot and Robert J. Deri.....................................................................................................1-1

Combined Sensing and Simulation for Enhanced Evaluation of StructuresDavid B. McCallen, Mathew Hoehler, Gregory A. Clark, and James V. Candy .............................................1-3

Modeling and Simulation of Wireless Sensor NetworksRowland R. Johnson ................................................................................................................................1-11

Information Warfare Analysis CapabilityJ. C. Smart ..............................................................................................................................................1-15

2. Center for Microtechnology

OverviewRaymond P. Mariella, Jr., Center Leader

Disposable Microfluidic Biological Sample Preparation SystemRobin R. Miles, Daniel L. Schumann, Kelye A. Allen, Jim A. Butler, and Kerry A. Bettencourt ...................2-1

Sub-Micron Lithography with a 5X StepperDino R. Ciarlo and Benjamin P. Law ..........................................................................................................2-5

Advanced Packaging for Wireless Microsensor ModulesAbraham P. Lee, Charles F. McConaghy, Jimmy C. Trevino, Leslie M. Jones, and Jonathan Simon.............2-7

Radio Frequency Technology for Wireless Microsensor ModulesCharles F. McConaghy, Abraham P. Lee, Charles Chien, Chris Deng, and Igor Elgorria ...........................2-11

Ultra-High-Speed Analog-to-Digital Conversion TechnologyMark E. Lowry, Ronald E. Haigh, and Charles F. McConaghy ..................................................................2-13

HV PhotovoltaicsKarla G. Hagans and Ronald E. Haigh......................................................................................................2-17

Contents

FY 98 i

Contents

Lattice Boltzmann Simulation of Complex Fluid FlowsDavid S. Clague .......................................................................................................................................2-19

Micro-Electromechanical Systems (MEMS) for Characterization of Plastic-Bonded ExplosivesJeffrey D. Morse, Dino R. Ciarlo, Scott E. Groves, Diane J. Chinn, Mehdi Balooch, and Mark J. LaChappel............................................................................................................................2-23

Micro-Electromechanical-Systems-(MEMS)-Based Fuel Cell TechnologyJeffrey D. Morse, Robert T. Graff, Alan F. Jankowski, and Jeffrey P. Hayes ..............................................2-27

High-Power GaN Microwave Device TechnologyGlenn A. Meyer, Gregory A. Cooper, Stacy L. Lehew, Thomas W. Sigmon, Daniel Toet, Steven DenBaars, and Umesh Mishra ......................................................................................................2-31

Glass EtchingHarold Ackler and Stefan P. Swierkowski.................................................................................................2-37

Micro-Electromechanical Systems Foundry InterfaceMichael D. Pocha ....................................................................................................................................2-41

A Hydrogel-Actuated Implantable SensorAmy W. Wang, Abraham P. Lee, Charles F. McConaghy, Christopher B. Darrow, Aleksandr Gilman, Stephen M. Lane, and Joe H. Satcher, Jr. ...............................................................................................2-45

Lambda Connect: Multi-Wavelength Technologies for Ultrascale ComputingRobert J. Deri, Michael C. Larson, Michael D. Pocha, and Mark E. Lowry...............................................2-49

3. Center for Precision Engineering

OverviewKenneth L. Blaedel, Center Leader

Micro-Drilling of ICF CapsulesSteven A. Jensen and Brent C. Stuart ........................................................................................................3-1

A Spatial-Frequency-Domain Approach to Designing Precision Machine ToolsDebra A. Krulewich ...................................................................................................................................3-3

Precision Grinding of Brittle MaterialsMark A. Piscotty, Kenneth L. Blaedel, Pete J. Davis, and Pete C. Dupuy....................................................3-9

Engineering Research Development and Technologyii

Contents

FY 98

4. Center for Computational Engineering

OverviewPeter J. Raboin and Clfiford C. Shang, Center Leaders

Hybrid Ray/Wave Methods for Optical PropagationRichard P. Ratowsky, Jeffrey S. Kallman, Michael D. Feit, and Bedros B. Afeyan........................................4-1

Technologies for Advanced Induction AcceleratorsMaurice A. Hernandez ...............................................................................................................................4-7

TIGER: An Object-Oriented Time-Domain Simulation CodeDavid J. Steich, Jeffrey S. Kallman, Gerald J. Burke, S Terry Brugger, and Daniel A. White.......................4-9

OPUS: An Optically Parallel Ultrasound SensorJeffrey S. Kallman, Dino R. Ciarlo, Elaine Ashby, and Graham H. Thomas ...............................................4-13

Optical Transition Radiation Diagnosis for Electron BeamsGregory P. Le Sage and Roger A. Richardson...........................................................................................4-19

Characterization of Electromagnetic Scattering from Defects in the EUVL ProcessLisa Wang, Scott D. Nelson, Jeffrey E. Mast, and Abbie L. Warrick..........................................................4-23

Nuclear and Electromagnetic Radiation Simulation Tools for Dual-Revalidation of the StockpileDavid J. Mayhall and Michael F. Bland.....................................................................................................4-29

Self-Effects in Expanding Electron Beam PlasmasManuel Garcia .........................................................................................................................................4-33

Pump-Induced Wavefront Distortion in Prototypical NIF and LMJ AmplifiersMark D. Rotter, Kenneth S. Jancaitis, Christopher D. Marshall, Luis E. Zapata, Alvin C. Erlandson, Geoffroy LeTouze, and Stephane Seznec ...................................................................4-37

Parallel Algorithm Development for Computational MechanicsCarol G. Hoover, Robert M. Ferencz, Anthony J. De Groot, Robert J. Sherwood, Edward Zywicz, Yuen L. Lee, and Douglas E. Speck.................................................................................4-47

DYNA3D-TOPAZ3D Coupling and DYNA3D-NIKE3D LinkageJerry I. Lin ..............................................................................................................................................4-55

A Physically Stabilized Eight-Node Hexahedral ElementMichael A. Puso.......................................................................................................................................4-59

A Cyclic Viscoplastic Constitutive ModelPhani Kumar V. V. Nukala ........................................................................................................................4-65

Electromagnetic Cold-Test Characterization of the Quad-Driven Stripline KickerScott D. Nelson and James E. Dunlap......................................................................................................4-71

Photonic Doppler VelocimetryPaul D. Sargis, Nicole E. Molau, and Daren Sweider................................................................................4-77

iii

Contents

Modeling Coupled Heat and Mass DiffusionArthur B. Shapiro and Philip M. Gresho...................................................................................................4-81

Analysis and Modeling of a Stripline Beam Kicker and SeptumBrian R. Poole, Lisa Wang, Yu Ju (Judy) Chen, and George J. Caporaso ..................................................4-85

5. Center for Nondestructive Characterization

OverviewHarry E. Martz, Center Leader

Techniques for Enhancing Laser Ultrasonic Nondestructive EvaluationGraham H. Thomas, Robert D. Huber, Diane J. Chinn, James V. Candy, and James Spicer ........................5-1

Nondestructive Evaluation of an Aluminum Alloy Using Hyperspectral Infrared Imaging Randy S. Roberts .......................................................................................................................................5-9

In-Situ Identification of Anti-Personnel Mines Using Acoustic Resonant SpectroscopyRandy S. Roberts and Roger L. Perry.......................................................................................................5-13

An Acoustic Technique for the Non-Invasive In-Situ Measurement of Crystal Size and SolutionConcentrationDiane J. Chinn, Paul R. Souza, and Harry F. Robey..................................................................................5-19

Micro X-Ray Computed Tomography for PBX CharacterizationDiane J. Chinn, Jerry J. Haskins, Clinton M. Logan, Dave L. Haupt, Scott E. Groves, John Kinney, and Amy Waters..................................................................................................................5-23

Evaluation of an Amorphous Selenium Array for Industrial X-Ray ImagingClinton M. Logan, Jerry J. Haskins, Kenneth E. Morales, Earl O. Updike, James M. Fugina, Anthony D. Lavietes, Daniel J. Schneberk, Gregory J. Schmid, Keo Springer, Peter Soltani, and Kenneth Swartz.................................................................................................................................5-27

LANDMARC Radar Mine DetectionStephen G. Azevedo, Jeffery E. Mast. James M. Brase, and E. Tom Rosenbury ........................................5-39

IMAN-3D: A Software Tool-Kit for 3-D Image AnalysisSailes K. Sengupta...................................................................................................................................5-51

Image Recovery Techniques for X-Ray Computed Tomography in Limited-Data EnvironmentsDennis M. Goodman, Jessie A. Jackson, Maurice B. Aufderheide, and Erik M. Johansson.......................5-61

Engineering Research Development and Technologyiv

Contents

FY 98

6. Supporting Technologies

OverviewRobert T. Langland

Modeling of Anistropic Inelastic BehaviorDaniel J. Nikkel, Jr., Arthur A. Brown, and James Casey ...........................................................................6-1

Modeling Large-Strain, High-Rate Deformation in MetalsDonald R. Lesuer, Mary M. LeBlanc, and Gregory J. Kay ...........................................................................6-7

Uniform Etching of 85-Cm-Diameter GratingSteven R. Bryan, Jr. and David L. Sanders...............................................................................................6-17

Distributed Sensor Inertial Measurement UnitCarlos A. Avalle and John I. Castor..........................................................................................................6-23

Fiber-Based Phase-Shifting Diffraction Interferometer for Measurement and Calibration of the Lick Adaptive Optics SystemEugene W. Campbell and Jong R. An .......................................................................................................6-27

Selected Engineering Publications

Selected Engineering Publications......................................................................................................PB-1

Author Index

Author Index...........................................................................................................................................AI-1

v

Spiros Dimolitsas, Associate Director for Engineering

Cultivating our new researchcenter organization

Nineteen ninety-eight has been a transition yearfor Engineering, as we have moved from our tradi-tional focus on thrust areas to a more focusedapproach with research centers. These five newcenters of excellence collectively compriseEngineering’s Science and Technology program. Thispublication summarizes our formative year underthis new structure.

Let me start by talking about the differencesbetween a thrust area and a research center. Thethrust area is more informal, combining an impor-tant technology with programmatic priorities. Incontrast, a research center is directly linked to anEngineering core technology. It is the purer model,for it is more enduring yet has the scope to be ableto adapt quickly to evolving programmatic priorities.

To put it another way, the mission of a thrust areawas often to grow the programs in conjunction witha technology, whereas the task of a research centeris to vigorously grow our core technologies. By culti-vating each core technology, we in turn enable long-term growth of new programs.

Taking a longer view

The emphasis in the new research center struc-ture is thus more long-term, building a research anddevelopment capability for the Laboratory for thefive-year timeframe; a thrust area was often moreshort-term and narrower in technical scope, concen-trating on the one- to two-year outlook of the tech-nological needs of the programs. Our new researchcenters are positioned primarily not to solve today’sproblems but to pioneer innovative R&D work foremerging problems.

In a sense, we are internal venture capitalists forour Lab programs. It’s in this way that we can helpthe Lab’s programs attract funding for the long run.

Engineering must work to sustain theLaboratory’s competitive advantage; we must createthings that are technically one-of-a-kind. Within ourorganization, this means we are able to capitalize on

our unique capabilities, specifically our comprehensiveapproach with dual disciplines. Many organizationshave strong mechanical engineering programs;others offer solid electronics engineering. We haveover a 40-year tradition of cross-pollinating both.

Our new centers are specifically designed tobring together the best of both disciplines, creatinga synergy that most organizations can’t. For exam-ple, as we model microscale systems, we advance amicrotechnology that is unique in how it incorpo-rates traditional electronic fabrication techniqueswith microminiature electromechanical designs.

For this reason, we have reorganized theEngineering Directorate to create a merged struc-ture: the Engineering Science and Technologyprogram, in addition to the Mechanical andElectronics Engineering programs.

The year ahead

Engineering is committed to the centers of excel-lence concept and will provide resources to seedbold new technologies. At the same time, becauseour new centers are closer to our core competen-cies, they will be easier to manage and offer moreflexibility. Already we can see results. This year,Engineering competed much more successfully forthe Laboratory’s LDRD investment dollars that fundinnovative projects.

In early 1999, we should complete the forma-tion of our five centers and select their long-termleadership. Our second objective for the new yearis synthesize our multiple visions for the centersso that we can write an updated version of ourstrategic plan.

Look through this report. It can be read as anoverview of our new research centers, their enablingtechnologies, and supporting capabilities—and whatwe achieved this year.

Our future depends on how we find innovative butcost-effective engineering solutions to emergingtechnical problems. Through our partnerships withDOE and UC and a new center-based structure, weare planting new ideas and cultivating solutions thatwill grow value on a national scale.

Introduction

FY98 Dividers 8/19/99 5:34 PM Page 2

Introduction


The Center for Complex Distributed Systems is afocus for advanced technology development andapplications involving ultrascale, spatially distrib-uted systems. The Center seeks to develop anadvanced suite of technical capabilities that enablethe engineering of such systems, and to providethese capabilities in support of missions andprograms at Lawrence Livermore NationalLaboratory (LLNL).

“Complex distributed systems” can be looselydefined as aggregations of large numbers of coupled,cooperating elements or subsystems, in which thesystem behavior cannot be described by simple hier-archical or nearest-neighbor interactions betweenelements. Examples include distributed sensornetworks and beam diagnostics, large-scale distrib-uted control and communication systems for opticaland/or accelerator beams, and distributed informa-tion processing systems based on an underlyingdistributed sensor or control system.

Complex distributed systems play a critical rolein several of LLNL’s national security missions, withthe potential to impact LLNL programs in otherareas as well. This is illustrated by the four projectsdescribed in this report, all of which address appli-cations of national importance and current interest:

1. “Accelerating the Development of PetaflopApplications and Systems” explores extremelylarge computational platforms, for applicationareas such as weapon simulations andclimate modeling.

2. “Combined Sensing and Simulation forEnhanced Evaluation of Structures” describesthe use of networked sensors to monitor therobustness of critical structural systems, suchas bridges, dams, and buildings.

3. “Modeling and Simulation of Wireless SensorNetworks” investigates wireless sensornetworks for applications such as early warningagainst attack by chemical agents. Thenetworked sensor technology described in thiswork, and in the preceding article, will alsoimpact environmental and industrial monitoring.

4. “Information Warfare Analysis Capability”discusses the challenges associated with, andtechniques required for, defending informationsystems against attack.

A wide variety of technical disciplines arerequired for ultrascale system engineering. At thistime, the Center is focussing on the areas of commu-nications and control, system engineering, and simu-lation. These areas present fundamental challengesfor most large-scale systems, as illustrated by allprojects described in this report.

Communication and data flow issues are centralto all four projects, as is the ability to simulate anddesign ultrascale systems. This underlying common-ality of basic issues indicates that the technologicalcapabilities developed by these investigators isgenerally applicable and transferable to other appli-cations for complex distributed systems.

Robert J. Deri, Center Leader

Center for Complex Distributed Systems


Center for ComplexDistributed Systems

1


Contents

1. Center for Complex Distributed Systems

OverviewRobert J. Deri, Center Leader

Accelerating the Development of Petaflop Applications and SystemsAnthony J. De Groot and Robert J. Deri.....................................................................................................1-1

Combined Sensing and Simulation for Enhanced Evaluation of StructuresDavid B. McCallen, Mathew Hoehler, Gregory A. Clark, and James V. Candy .............................................1-3

Modeling and Simulation of Wireless Sensor NetworksRowland R. Johnson ................................................................................................................................1-11

Information Warfare Analysis CapabilityJ. C. Smart ..............................................................................................................................................1-15

Engineering Research Development and Technology

ccelerating the Development of Petaflop Applications and Systems


Introduction

The purpose of this project is to accelerate thedevelopment of technologies necessary to thecreation of petaflop scientific applications andpetaflop computing systems. This project bringstogether a multi-disciplinary team at LawrenceLivermore National Laboratory (LLNL) (computa-tions, defense, engineering, physics, and lasers) todevelop a modeling and analysis framework that willpredict performance and identify bottlenecks forhighly-relevant, unclassified Accelerated StrategicComputing Initiative (ASCI) applications (scaled topetaflop-class) running on proposed commodity-based architecture designs.

This will give LLNL a new capability that couldassist in developing new high-performance applica-tions, evaluating new hardware acquisitions, andcreating a balanced computing environment.

Progress

One of the activities for this fiscal year was tobegin development of a simulation capability tounderstand how interprocessor communicationaffects performance.

Simulation of the execution of large, complexapplications on petaflop computing systems is acomputationally intensive task. In the past, we havesimulated the execution of an application at theinstruction level. The computation required for thishighly detailed simulation is approximately onehundred times more than the computation time to

run the application on a real machine. We chosetrace-based simulation to make this problemcomputationally tractable.

In our trace-based simulation, the applicationwith its input problem are executed on a hostmachine, producing a trace of events. Eventsinclude all message passing operations, andoptional user-defined events, including key subrou-tine entrances and exits. This trace, along with adescription of the computer architecture to besimulated is used as input to a discrete-eventsimulator, that simulates the execution of theapplication on the simulated architecture.

Generation and visualization of traces andsubsequent analysis have already improved perfor-mance of a complex application. We chose ParallelDYNA3D (ParaDyn) as the first application to studybecause it is one of the few parallel productioncodes far enough along in development for thisproject to make an immediate impact. We used theAIMS code from NASA Ames to generate anddisplay the event traces from ParaDyn execution.

Figure 1 shows the execution of ParaDyn oneight processors. The problem simulated byParaDyn is that of a buckling beam. In the figure,the horizontal direction represents the passage oftime, and the vertical direction represents processornumber. Line segments connecting bars representmessage communication from the correspondingprocessors. The shading of a bar represents theexecution of a set of subroutines by the correspond-ing processor. Figure 1 shows that most of the timeis used for computation and very little is used for

FY 98 1-1

We are creating a modeling and analysis framework to accelerate the development of petaflopapplications and systems.

Anthony J. De GrootDefense Sciences Engineering Division Electronics Engineering

Robert J. DeriElectronics Engineering Technologies DivisionElectronics Engineering

(7)

(6)

(5)

(4)

(3)

(2)

(1)

(0)


communication. The figure also shows that theexecution of the subroutine represented by the blackbars is excessively long.

Further examination of the code showed that anoptimization could be made to that subroutine to

reduce execution time. Figure 2 shows the result ofthat optimization, where the total execution time ofthat ParaDyn problem was reduced by a factor ofabout 1.5. Further optimizations may be possiblewith additional study.

Engineering Research Development and Technology1-2

Figure 1.Visualization ofParaDyn executiontrace, showing excessive time spentin the subroutine,represented by theblack bars.

(7)

(6)

(5)

(4)

(3)

(2)

(1)

(0)

Figure 2.Visualization ofParaDyn executiontrace after programoptimizationsuggested by Figure 1.

ombined Sensing and Simulation for Enhanced Evaluation of Structures


Introduction

With the proliferation and advancement of sophis-ticated numerical simulation software tools over thepast twenty years, computational simulation of largestructural systems has been a subject area experi-encing rapid growth. Engineers now rely heavily onlarge-scale structural simulations to design andevaluate the performance of critical new structures,and to establish the performance of expensive retro-fits on existing structures.

Despite the advances in computational methods,there remains a significant degree of uncertainty inpredicting the field performance of many large-scalestructural systems. These uncertainties are rootedin our inability to precisely quantify the phenomeno-logical behavior of certain aspects of structural exci-tation and structural response—for example, uncer-tainties in the precise deformation characteristics ofcomplex structural element interconnections oruncertainties in estimation of the actual loads astructure will be subjected to. To advance our abilityto accurately and confidently simulate the response

of structures, we must make use of measuredstructural response characteristics and measuredexcitation functions.

The notion of coupling simulation and measure-ment is emerging in the study and characterizationof many complex systems found in the sciences andengineering. One of the tenets of a new NationalScience Foundation program on Knowledge andDistributed Intelligence is the recognition that:“Better understanding of complex phenomena nowrequires interplay between computations and data,often in real time.” The work described in this reportprovides the foundation at LLNL for establishing aninterplay between computations and measured datafor the specific case of large structural systems.

The essential links between simulation andmeasurement are shown in Fig. 1. As indicated,information from a numerical structural model anddata from field measurements of structural responseare fed into a model-based signal processor. Thesignal processing toolbox is used to evaluatewhether the model and as-built structure are inagreement, or if there is a discrepancy between the

FY 98 1-3

Large-scale computer simulation is an essential tool in the design and analysis of modern struc-tures. With the enormous cost and construction effort associated with many large structures, it isimperative that computer simulations provide an accurate picture of structural performance span-ning a multiplicity of loading environments such as bomb blasts, earthquakes, and ambient vibra-tions. In the current study, techniques are investigated that allow evaluation of simulation modelaccuracy and the possibility for subsequent enhancement. The multidisciplinary tools being used atLawrence Livermore National Laboratory (LLNL) include finite-element-based structural simulation,model-based signal processing, and remote sensing and data communication. The overall objective ofthis research is to symbiotically couple numerical simulation with field measurement of structuralbehavior. The result will be enhanced accuracy and reliability of numerical simulation of structuralresponse, and the ability to monitor fundamental changes in complex structural systems, a prerequi-site to health monitoring and damage detection.

David B. McCallen and Matthew HoehlerNew Technologies Engineering DivisionMechanical Engineering

Gregory A. Clark and James V. CandyElectronics Engineering Technologies DivisionElectronics Engineering


dynamical behavior of the model and that of theactual structure.

In addition, research on the signal processingtoolbox will attempt to discriminate the source ofexisting discrepancies between the simulation modeland the as-built structure. Once discrepancies canbe identified, the basis for damage detection in astructural system has been established. Experimentalmeasurements before and after an event can beused to identify changes in the structural system,and identification of the source of the differenceswith a damage detector can be performed to investi-gate where damage has occurred in a largedistributed structure.

Progress

In the first year of work, we have assembled amultidisciplinary team, the combined expertise ofwhich spans the required technologies in mechani-cal and electronics engineering. Our accomplish-ments include the following:

1. An understanding of the relationship betweenthe structural simulation models of mechani-cal engineering and the state space models ofelectronics engineering has been establishedfor large structures.

2. The protocols and requirements for informa-tion exchange between second order equationsof motion of a computational structural model,and first order equations of motion of statespace, have been developed.

3. The software for a model-based signalprocessing state space model module hasbeen completed for performance of structuralsimulations, and a state space filter has beendeveloped for detection of differencesbetween the simulation model and ameasured structural response.

4. In the area of sensing and communication, anew data acquisition system has been devel-oped which will allow on-demand wirelesscommunication of data via cellular phone froman instrumented structure.


Measured forcing function

Measured response from distributed sensors

3) Model based signal processor

Finite element basednumerical model

2) Numerical simulation modeli) Simulator

ii) Difference detector

iii) Difference identifier

Filter/ID

Systemmatrices

Information feed back loop

Damage assessment

Identify discrepancies betweenmodel and actual structure

Characterize the natureof the discrepancies

1) Measurement and communication

Measured input and response

State space model

State space filter

[M][K][C]

M[ ] x C[ ] x K[ ] x + + f t( ) =

Figure 1. Model-based signal processing for linkage betweennumerical modelsand measuredresponse.


5. A field experiment has been completed at theNevada Test Site to evaluate the data acquisitionsystem and to provide a real world data set forevaluation of the model-based signal processingalgorithms in year two of this project.

Interfacing Numerical Simulation Modelsand Signal Processing

A capability for the model-based signal process-ing algorithms to detect discrepancies between anumerical structural model and the actual structure

has been developed and coded in a MATLAB environ-ment (Fig. 2).1 The matrices constructed for thenumerical simulation model are passed into theMATLAB-based state space model for process simu-lation (essentially a recast of the “N” second orderequations of motion into a set of “2N” first orderequations). Using the constructed state space modeland measured data from a structure, a differencedetector based on the whiteness test described byCandy2 has been coded. This whiteness test, basedon an autocorrelation check between measuredstructural response and the state space model,

FY 98 1-5

Numerical simulation model

“N” 2nd order ordinary differential equations

M[ ] C[ ] K[ ], ,f(t) d(t)

Input forcingfunction

Output response

Signal processing

Finite element simulation model

“2N” 1st order ordinary differential equations

w t( ) v t( )

f(t) d(t)

State space simulatorInput forcingfunction

Output response

A M[ ] C[ ] K[ ], ,( )[ ]

B M[ ]( )[ ]

H

A[ ]0[ ] I[ ]

M[ ]1–

K[ ]–[ ] M[ ]1–

C[ ]–[ ]= B[ ]

0[ ]

M[ ]1–

=

Z t( ) 0

T1

T d t( ) =

A M[ ] C[ ] K[ ], ,( )[ ]

B M[ ]( )[ ] H

Rww

Rvv

,( ) f

Z t( ) X t( ) Z t( )

X t( ) d t( )

d'ˆ t( ) = Z t( )

0 T

1 T d t( ) =

State space filter(difference detector)

Figure 2. Numericalsimulation modelmatrix hand-off tosignal processingstate space simulatorand state space filter.


provides a rapid assessment of the validity of thenumerical structural model.

An important feature of the model-based signalprocessor under development is the ability to workwith full transient time history waveforms of thestructural response. This is in contrast to mostestablished model evaluation algorithms which arebased on natural modeshape characteristics of thestructure.3 The ability to work directly with tran-sient time histories will provide a powerful toolwhich can use any forcing function on the structure(that is, an impulsive load or sinusoidal loadimposed on the structure by design). Furthermore,the signal processing methodology is not restrictedto linear systems.

The ability of the signal processing algorithms todetect differences between the numerical struc-tural model and the actual structure has beendemonstrated with example problems. Figure 3illustrates an example for a five-story building, inwhich a simple five-degree-of-freedom “shear build-ing” numerical simulation model of the structure hasbeen constructed. In addition, a second model isconstructed to be representative of the data from anactual “measured” structure. The second model isintended to provide the data which typically wouldbe measured in the field.

To evaluate the ability of the signal processingalgorithms to detect the difference between the modeland the “measured” structure, transient responsewaveforms were determined for unit forcing functionsapplied to each floor level of the sample building.When the simulation model and “measured” structureare precisely the same structure, the whiteness testwhich discriminates differences between the modeland actual structure is passed, indicating the model isin agreement with the actual structure (Fig. 3a).However, when the model representing the measuredstructure is perturbed by cutting the stiffness of thefirst floor level by one-half (Fig. 3b), the signalprocessing package immediately senses the differencebetween the numerical simulation model and the“measured” structure, providing the engineer with animmediate indicator of the deficiencies of the numeri-cal structural model.

Sensing, Communication, and Field Experimentation

The signal processing linkage between thenumerical structural model and the measured struc-ture requires the ability to economically monitor theresponse of a large structural system. For manylarge, distributed structures this can be a difficult,

and potentially prohibitive task if traditional wiredsensor systems are used. In cooperation with theprivate firm of Jarpe Data Systems, a data acquisi-tion system has been developed that will allowremote gathering of transient response data from alarge structure. The capabilities of this data acquisi-tion system have been tested in an experiment at theNevada Test Site (Fig. 4). The data acquisitionsystem records a data stream from a suite of struc-tural sensors, typically accelerometers, and storesthe time-stamped data on a disk storage system. Theacquisition system has a cellular phone system onboard and can be called up remotely with a laptopcomputer with a modem for on-demand download ofdata. This data acquisition system will provide aneconomical and practical means of remotely moni-toring large distributed structures.

Future Work

The development of the model-based signalprocessing toolbox will continue with the construc-tion of the Difference Identifier (the Filter/ID inFig. 1). This will complete the basic signal process-ing package. Extensive sensitivity studies will beperformed to assess the ability of the signalprocessing tools to detect and identify many differ-ent types of discrepancies between simulationmodels and measured structures. These studies willmake use of computer simulation models (moresophisticated samples of the type shown in Fig. 3),existing full-scale structural tests performed atvarious universities, and the data obtained from thecarefully controlled structural experimentperformed at the Nevada Test Site in FY-98. In addi-tion to identification of discrepancies, the researchwill determine the degree to which the precisecause of the discrepancies can be identified by theDifference Detector.

In parallel to the algorithmic developments,sensors and data acquisition systems will be placedon three large distributed structures to establish theability to perform on-demand remote sensing overan extended time period. The target structuresinclude the San Francisco-Oakland Bay Bridge4,5

the National Ignition Facility laser bay structures,and the Bixby Creek Arch Bridge at Big Sur(Fig. 5).6

Each of these structures has been studied atLLNL as part of ongoing research and develop-ment projects or programmatic work, andnumerical simulation models exist for eachstructure. These important structures presentdifferent sensing and communication challenges



FY 98 1-7

[K]

K1+K2 –K2 0 0 0

–K2 K2+K3 –K3 0 0

0 –K3 K3+K4 –K4 0

0 0 –K4 K4+K5 –K5

0 0 0 –K5 K5

=

System stiffness matrix(from structural model) i) Simulator ii) Difference

detector

State spacefilter

State spacemodel

P5(t)

Simulationmodel

“Measuredstructure”

Model #1

(a)

(b)

Model #2

P1(t)

P2(t)

P3(t)

P4(t)

M5

M1

M2

M3

M4

M5

M1

M2

M3

M4

K5

K1

K2

K3

K4

K5

K1

K2

K3

K4

P5(t)

P1(t)

P2(t)

P3(t)

P4(t)

P5(t)

P1(t)

P2(t)

P3(t)

P4(t)

M5

M1

M2

M3

M4

M5

M1

M2

M3

M4

K5

K1

K2

K3

K4

K5

K1/2

K2

K3

K4

P5(t)

P1(t)

P2(t)

P3(t)

P4(t)

Transient response Whiteness test (passes)

Simulationmodel

“Measuredstructure”

Transient response Whiteness test (fails)

Figure 3. Development of model-based signal processing components—a model-based simulator for process simulation and a statespace filter for difference detection. a) Application to a problem in which the computational model and measured structure agree(whiteness test is passed); b) application to a problem in which the computational model and measured structure disagree (whiteness test fails).

and will require structure-specific detailing ofthe sensors and data acquisit ion system.Previous study of these structures has providedthe understanding of the frequency range ofinterest, and existing simulation models will beused to establish sensor placement locations.Successful monitoring of these structures willprovide the hardware and methodology for

deployment of dense instrumentation arrays onlarge distributed structures.

Large structural systems such as the NationalIgnition Facility or the Bay Bridge represent a tremen-dous capital investment and are critical structures forour society. The degree to which we analyze, designand monitor these structures should be commensu-rate with the cost and criticality of function which



Single-boardcomputer

GPS

A/Dconverter

Cellularphone

modem

Diskdrive

Mode #1 f=5.8 Hz Mode #3 f=7.5 Hz Mode #4 f=18.5 Hz

Precision timing Remote communication

Data storage

Figure 4. Data acquisition system for remote, wireless monitoring of a distributed structure. System includes GPS for precision time-stamping of data, and on-board cellular phone communications.


FY 98 1-9

Simulation model from Caltranssponsored research project

Simulation modelfrom LLNL/UC CLCresearch project

Large distributed structurewith very broad band frequencycharacteristics, requires strong and weak motion measurements

Large distributed structurewith small amplitude, highfrequency content motions

Remotely located structurewith severe accessibilitylimitations

Simulation modelsfrom programmaticwork

Figure 5. Large structural systems to be instrumented in FY-99.

they perform. The methodologies developed withthis research will provide the tools necessary tosignificantly improve our understanding of howlarge distributed structures behave, will lead to

increased reliability in numerical simulations,and will provide an entirely new capability forlong term monitoring to ensure the integrity ofimportant structures.


References

1. Clark, G. A., “Failure Detection for MechanicalStructures: A State-Space Model Base Approach,”Lawrence Livermore National Laboratory Report,in preparation.

2. Candy, J. V., and D. L. Lager (1979),“Identification, Detection and Validation ofVibrating Structures: A Signal ProcessingApproach,” Lawrence Livermore NationalLaboratory, Livermore, California (UCID-18720).

3. Berman, A., and E. J. Nagy (1983), “Improvement of aLarge Analytical Model Using Test Data,” AIAAJournal, Vol. 21, No. 8, August.

4. McCallen, D. B., and A. Astaneh-Asl (1988),“Computational Simulation of the NonlinearResponse of Suspension Bridges,” Proceedings ofthe World Structures Congress, Elsevier, San Francisco, California.

5. McCallen, D. B., and A. Astaneh-Asl, “Explicit DynamicAnalysis of Cable Supported Bridges,” in preparation,to be submitted to Computers and Structures.

6. McCallen, D. B., C. R. Noble, M. S. Hoehler, and M.Gerhard, “The Seismic Response of Concrete ArchBridges with Focus on the Bixby Creek Bridge,Carmel, California,” Lawrence Livermore NationalLaboratory Report to the California Department ofTransportation, in preparation.


odeling and Simulation of Wireless Sensor Networks


Introduction

Recent advances in micro-electromechanicalsystems, low-powered circuit technologies, andcomputer networking have made possible a low-powered WSN involving thousands or millions ofsensor nodes. Over the next ten to twenty years,LLNL will have the opportunity and obligation toparticipate in programs involving these networks.Just as the capability to model and simulate nuclearexplosions was an R&D enhancer for nucleardevices, the capability to model and simulate WSNswill be an R&D enhancer for WSNs. A modeling andsimulation capability 1) will dramatically reduce theeffort required to develop a WSN; 2) can be used asa planning tool for a deployment; and 3) can be usedto evaluate proposed network systems.

Current wireless network techniques do notmeet node power consumption and total systemthroughput requirements. Academic researchersare beginning to consider non-hierarchical networktopologies as a means to address power consump-tion and system bandwidth issues. Data packetsare routed from node to node based on which RFchannel offers the most efficiency at a particulartime (usually, this results in a data packet beingrouted to a nearby node). Higher RF channel effi-ciency allows a lower RF power level to be used,which reduces the node power consumption. It alsoreduces RF interference by other nodes, therebyincreasing total system throughput. It has beenshown that, although latency usually increases(due to increased hop count), power consumptiondecreases and system throughput increases.

The disadvantage of non-hierarchical networktopologies is that their behavior is only beginning tobe understood. In addition, research has beenfocused mainly on applying these techniques topersonal communication systems, such as cellulartelephones and personal digital assistants. It isunknown how effective these existing techniques willbe for the typical LLNL WSN application.

Progress

During FY-98 a core competency in modeling andsimulation techniques for WSNs was developed. Ateam of LLNL researchers have attained expertisein analyzing WSNs via modeling and simulationtechniques. Expertise has been developed in the useof state-of-the-art, commercially available simula-tion packages. In addition, researchers have devel-oped key computer software components. Wirelesssensor networks being considered are tightlycoupled with a physical system that will affect theiroperation. Therefore, it is important to be able tointegrate a simulator of such a physical system withthe WSN simulation.

Consider, for example, a chemical attack warningsystem. Terrain and atmospheric conditions willaffect both the movement and dispersion of a chemi-cal release, and the ability of the wireless nodes tocommunicate with one another. An integrated simu-lation system makes it possible to investigate systemissues such as the effect of terrain on battery life.

The development of the core competency hasbeen guided by the requirements of two existingWSN programs at LLNL: the Joint Biological Remote

FY 98 1-11

Currently, Lawrence Livermore National Laboratory (LLNL) has several wireless sensor network(WSN) programs, and it is likely that the number will increase. In FY-98, LLNL researchers developeda core competency to model and simulate WSNs, which will give LLNL a competitive advantage inresearch and development. Contributions to existing programs are described, and the results ofLLNL-funded academic research are presented.

Rowland R. JohnsonComputer and Communications Engineering DivisionElectronics Engineering


Early Warning System (JBREWS) and the WirelessSensor project.

JBREWS will consist of 100 to 1000 biologicalsensor nodes, each capable of wireless communica-tion with other sensor nodes or a central controlnode. The objective is to detect, track, and predictthe movement of airborne biological agents.JBREWS is a multi-phase program, each phasebased on newer, more sophisticated technology.

The performance of a JBREWS deployment willdepend on sensor node placement, terrain, andatmospheric conditions. An overall system effective-ness is determined by obtaining performance crite-ria for widely varying deployment scenarios.Obtaining performance criteria for an actual deploy-ment is expensive, and simulation offers a cost-effective alternative.

A primary concern of the JBREWS project is thatthe system be able to configure itself (Fig. 1) after

sensor nodes have been randomly deployed, forexample, by being air-dropped. In this case it isimpossible to determine, a priori, what communica-tion links between the nodes will be available. Onceplaced, each sensor node must be able to determineother sensor nodes that it can communicate with.Then, the sensor nodes, acting collectively, mustconfigure themselves into a network. Several self-configuration algorithms were investigated.

JBREWS uses time-division multiple-access(TDMA) transport protocol as the basis for thecommunication channels. TDMA is characterized bya longer, though constant, latency when comparedto other transport protocols. By eliminating theacknowledgment of successful data packet receipt,the latency will be reduced. The attendant reductionin reliability is dependent on the terrain and theexact placement of sensor nodes. Investigation ofthis issue by analytic means is not feasible.


Figure 1. The JBREWSwireless sensornetwork configuringitself. The masternode has determinedthat it cannotcommunicate withslave_1 and slave_6(because of theobstruction).Therefore, it relies onslave_5 to relay datapackets to slave_1and slave_6.


However, simulation results have shown that theelimination of acknowledgment messages has hadthe unforeseen effect of reducing average latency,while increasing the variance in latency times.

The Wireless Sensor project is a joint projectbetween LLNL’s Microtechnology Center (MTC) andthe University of California, Los Angeles (UCLA) todevelop very low power wireless sensor nodes. Thecurrent focus of the project is on hardware develop-ment. In the next two or three years there will beopportunities to configure these nodes into systemsfor applications that are currently unknown. A simu-lation that accurately reflects current technology ortechnology being developed will provide the meansto explore new applications.

The key issue that will drive the design of asystem is the expected battery life (Fig. 2). This hasa direct impact on system timing, because accurateclocks use more power than less accurate clocks. In

turn, clock drift can degrade overall system perfor-mance by causing packet collisions and/or highlatency times. The current simulation accuratelyreflects 1) the power cost of computing and commu-nication operations, and 2) clock drift and systemtiming considerations.

In addition to work at LLNL, research at theUniversity of California, Berkeley (UCB) on efficientmulti-cast protocols was funded in FY-98. In manyWSN applications there is a control node that willsend directives to all sensor nodes. Typically,receipt of multi-cast packets is not acknowledgeddue to the network congestion that would occurwhen all sensor nodes acknowledge receipt at thesame time. Without acknowledgment, totally reli-able directive delivery is not possible, although it ispossible to achieve some degree of reliability.

The research task undertaken by UCB was todevelop methods to achieve high reliability with

FY 98 1-13

3.6000

3.5999

3.5998

3.5997

3.5996

3.5995

3.5994

3.5993

3.59920 10 20 30 40 50

Time (s)

0.30

0.25

0.20

0.15

0.0

0.05

0

-0.05

-0.15

-0.10

0 10 20 30 40 50Time (s)

Figure 2. Battery level and packet arrival times. (a) The battery is being drained whenever the receiver is turned on or the transmitteris used to transmit a packet; (b) shows a window in time when the receiver is turned on, to acquire packets being sent to it. To savepower the receiver is turned off outside this window. A clock drift is shown by the fact that packets arrive earlier and earlier untilfinally a packet arrival falls outside the window. At this point data packets are being lost until the system can re-sync itself.


non-acknowledged multi-cast protocols. Severalmulti-cast protocols were considered with theeffectiveness of each being determined by modelingand simulation.

The result of this research is a “tunable” multi-cast protocol, that is, by specifying the degree of reli-ability, protocol parameters are selected that yieldthe desired reliability with the least overhead.1,2

Future Work

Thus far, we have used commercially availablesimulation systems for the basis of our work. Thedisadvantage of this approach is that the simulationkernel can not be modified for the purpose of1) achieving tighter integration with existing

simulations of physical systems; or 2) porting the simu-lation kernel to a massively parallel computer. In FY-99we will develop the capability to construct simulationkernels that are specific to a particular application.

References

1. Wong, T., T. Henderson, S. Raman, and R. Katz(1998), “Exploiting Applications Level Framing inReliable Multicast for Periodic InformationDissemination,” Technical Final Report, University ofCalifornia, Berkeley, August.

2. Wong, T., T. Henderson, S. Raman, A. Costello,and R. Katz (1998), “Policy-Based TunableReliable Multicast for Periodic InformationDissemination,” Technical Final Report,University of California, Berkeley.


nformation Warfare Analysis Capability


Introduction

With the rapid growth of global computing andcommunications, information security is a criticalissue in all discussions of protection of the nationalinfrastructure. The purpose of our project—theInformation Operations, Warfare, and Assurance(IOWA) initiative—is to advance the enabling coretechnologies of this field.

Special emphasis was placed on computernetworks and telecommunications systems.

Progress

During FY-98, we developed 1) techniques foridentifying the topology of large, complex computernetworks; 2) data representation models for thesesystems; 3) high-performance methods for visualiz-ing the resulting complex models; 4) automatedanalysis methods for processing large-networkrepresentations; 5) specialized search techniquesfor isolating vulnerabilities; 6) a foundation forsimulating network operation; and 7) an assessmentmethodology for determining the consequences ofsystem component failure or disruption.

To automate information system protection, it isnecessary to first identify the visible componentsthat an intruder might attempt to access, and todetermine the specific information that might beinferred about each component.

We began by developing a set of softwaremodules for analyzing Internet-related protocols.This software examines the information that flowsacross a computer network and extracts networktopology and details about the configuration of eachcomponent. At present, the tool suite processes over

20 popular Internet protocols, retrieving over 50different system operating parameters.

Since modern computer networks have grownconsiderably in size (that is, more than 25,000nodes), a special model was developed to capturethe enormous amount of information that the toolsprocess. The resulting model uses a unique blend ofrelational database technology integrated into agraphical theoretical framework, providing rapidinformation retrieval in an environment conducive tolarge-network mapping and analysis.

We demonstrated a platform-independent viewerfor browsing the graphics model with integratedaccess to the relation database. To better managethe complexity of large networks, several powerfuldependency constructs, graphics operations, andreduction functions were incorporated into themodel. A diverse suite of generic graph-, fault-tree-,and Internet-specific processing algorithms wasdeveloped and demonstrated.

To better understand the nature of computer andnetwork vulnerabilities, a taxonomy of knownvulnerabilities was developed that forms the basis ofour new vulnerability database. This database wassubsequently populated with vulnerability facts fromindustry and private sources. The end result is a toolthat can now be used to automate the search forweaknesses in our computer systems.

Working with the modeling tools, an environmentwas constructed to perform high-fidelity simulationsof computer networks. The resulting tools can beused to simulate computer networks captured in theIOWA model. Arbitrary computer networks can alsobe constructed in the simulation environment andused to generate network traffic to test and cali-brate the network mapping tools.

FY 98 1-15

The Information Operations, Warfare, and Assurance initiative at Lawrence Livermore National Laboratoryhas advanced the enabling core technologies for information operations analysis. Special emphasis was placedon computer networks and telecommunications systems.

J. C. SmartElectronics Engineering Technologies DivisionElectronics Engineering

Raymond P. Mariella, Jr., Center Leader

Center for Microtechnology

In January of 1998, we took our latest PCRinstrument to the Joint Field Trials IV (JFT IV) inDugway, Utah, for the open competition and demon-strated the highest sensitivity ever recorded there fordetection of the plague simulant (Erwinia herbicola).Our microtechnology project on the fabrication offacets for GaN lasers has also shown excellentresults.

As the applications for microtechnology grow, wecontinue to perform the technology-base activitiesthat are needed to fulfill the needs of our LLNL andexternal partners and stay in a leadership position,both nationally and world-wide.

Overall, the numerous small investments inmicrotechnology have paid off this year in 1) outstanding results at JFT IV with the LLNL PCRinstrument; 2) the highest performance hand-heldgas chromatograph built, including microfabricatedinjector, column, and detector; 3) an LLNL elec-trophoresis system with etched/bonded platescontaining 384 channels; 4) $1M in DARPA fundingfor dielectrophoresis technology; 5) a $1.7MCRADA: a medical catheter device to releaseembolic coils into cerebral aneurysms that useslaser light for release and detection; 6) projects foroptical interconnects in partnership with thePhotonics Program, resulting in significant demon-stration of multiple-wavelengths carried per fiber;7) highly visible multi-million-dollar CRADA withsemiconductor businesses for EUV lithography; and8) collaboration with the University of California,Santa Barbara, on high-performance laser facetsfor GaN and demonstration of an advanced processin laser-based doping of GaN.

The mission of the Microtechnology Center atLawrence Livermore National Laboratory (LLNL) isto invent, develop, and apply microtechnologies inconjunction with programs in global security, globalecology, and bioscience.

Our capabilities cover materials, fabrication,devices, instruments, or systems that require micro-fabricated components, including micro-electro-mechanical systems (MEMS), electronics, photonics,microstructures, and microactuators. All of ourmicrotechnology work revolves around our micro-fabrication facility, and is driven principally by theapplications of our internal programs, and to alesser extent, by external applications. For both ofthese we must have multidisciplinary teams todeliver complete solutions to the problems.

The Microtechnology Center continued to grow inFY-98. Its more than 60 people have training in elec-tronics engineering, mechanical engineering, chemi-cal engineering, chemistry, physics, and thebiosciences. Our recent successes in analyticalinstrumentation reflect our broad, multidisciplinarybase and the cross-fertilization that results frompersonnel sharing their capabilities and ideas witheach other.

We continue to show a very high rate of return oninvestment. Over the last year, with a total budget ofmore than $15M, including approximately $0.6M oftechnology-base projects, we have successfullyperformed collaborations with industry, and haveachieved considerable national recognition. We havehad success in our on-going instrumentation project,developing both polymerase chain reaction (PCR)assays and instrumentation, supported by theDefense Intelligence Agency.



2


Contents

2. Center for Microtechnology

OverviewRaymond P. Mariella, Jr., Center Leader

Disposable Microfluidic Biological Sample Preparation SystemRobin R. Miles, Daniel L. Schumann, Kelye A. Allen, Jim A. Butler, and Kerry A. Bettencourt ...................2-1

Sub-Micron Lithography with a 5X StepperDino R. Ciarlo and Benjamin P. Law ..........................................................................................................2-5

Advanced Packaging for Wireless Microsensor ModulesAbraham P. Lee, Charles F. McConaghy, Jimmy C. Trevino, Leslie M. Jones, and Jonathan Simon.............2-7

Radio Frequency Technology for Wireless Microsensor ModulesCharles F. McConaghy, Abraham P. Lee, Charles Chien, Chris Deng, and Igor Elgorria ...........................2-11

Ultra-High-Speed Analog-to-Digital Conversion TechnologyMark E. Lowry, Ronald E. Haigh, and Charles F. McConaghy ..................................................................2-13

HV PhotovoltaicsKarla G. Hagans and Ronald E. Haigh......................................................................................................2-17

Lattice Boltzmann Simulation of Complex Fluid FlowsDavid S. Clague .......................................................................................................................................2-19

Micro-Electromechanical Systems (MEMS) for Characterization of Plastic-Bonded ExplosivesJeffrey D. Morse, Dino R. Ciarlo, Scott E. Groves, Diane J. Chinn, Mehdi Balooch, and Mark J. LaChappel............................................................................................................................2-23

Micro-Electromechanical-Systems-(MEMS)-Based Fuel Cell TechnologyJeffrey D. Morse, Robert T. Graff, Alan F. Jankowski, and Jeffrey P. Hayes ..............................................2-27

High-Power GaN Microwave Device TechnologyGlenn A. Meyer, Gregory A. Cooper, Stacy L. Lehew, Thomas W. Sigmon, Daniel Toet, Steven DenBaars, and Umesh Mishra ......................................................................................................2-31

Glass EtchingHarold Ackler and Stefan P. Swierkowski.................................................................................................2-37

Micro-Electromechanical Systems Foundry InterfaceMichael D. Pocha ....................................................................................................................................2-41

A Hydrogel-Actuated Implantable SensorAmy W. Wang, Abraham P. Lee, Charles F. McConaghy, Christopher B. Darrow, Aleksandr Gilman, Stephen M. Lane, and Joe H. Satcher, Jr. ...............................................................................................2-45

Lambda Connect: Multi-Wavelength Technologies for Ultrascale ComputingRobert J. Deri, Michael C. Larson, Michael D. Pocha, and Mark E. Lowry...............................................2-49


isposable Microfluidic Biological Sample Preparation System


Introduction

The purpose of this project was to develop amicrofluidic circuit based on thin film plastic. Fluidicchannels and plenums were formed using thin filmplastic such that they could be manipulated exter-nally to provide valving, pumping, and mixing func-tions, as shown in Fig. 1. The device can be used ineither a disposable or a semi-fixed system.

The chief advantage to such a system is that itprovides isolation of the biological fluid from theactuation device. Low-power actuators, such aspiezoelectric or electrostatic actuators, are gener-ally incompatible with fluid environments. High-workactuators, such as thermal actuators, require signifi-cant increases in power when in contact with fluids.

Other advantages to this system over alternativemicrofluidic approaches include the use of a singlematerial to reduce material compatibility concerns,reduce system cost, and minimize dead volume.

Progress

The primary focus of this effort was to develop afabrication process for defining and sealing theplastic sheets. A second effort was to develop an

external pneumatic actuator for the system toprovide valving, pumping, and mixing operations.

Forming the Plastic Sheet

Forming and bonding the plastic sheets proved tobe a difficult task. Our technique for manufacturingthe plastic device was to first pre-form the cavitieson each half of the structure, then bond the twohalves together to make the final structure. Wedeveloped several techniques to form the plasticstructure including heat-staking, laser welding, andmolding. Stiff plastics such as polypropylene andpolyethylene were first embossed to define thetubing and plenum areas of the structure, then laserwelded to a second sheet of plastic to complete thestructure. Elastomeric materials such as metal-locene and silicone were molded or heat-staked.

Thermal Heat-Staking. Several outside vendorswere contacted to provide heat-staking fixtures forfabricating the plastic structures. One vendor built afixture for bonding pre-embossed polypropylenesheets. Unfortunately, their fabrication techniqueproved unsuccessful as they melted the entire plas-tic structure into one big lump. A second vendorsuccessfully bonded two sheets of metallocene in

FY 98 2-1

We have fabricated plastic microfluidic devices to provide an inexpensive flow path for biologicalapplications. External actuation was used for valving, pumping, and mixing functions. The advantagesto this system over alternative microfluidic approaches are that it provides isolation of the biologicalmaterial from actuators, reduces material compatibility concerns and system cost, and minimizesdead volume.

Robin R. MilesElectronics Engineering Technologies DivisionElectronics Engineering

Daniel L. Schumann, Kelye A. Allen, and Jim A. ButlerManufacturing and Materials Engineering DivisionMechanical Engineering

Kerry A. BettencourtMaterials Science and Technology DivisionChemistry and Materials Science

205 Miles_qk 7/22/99 1:33 PM Page 2-1


the desired configuration to provide for the desiredinternal tubes and plenums. The bonds were capableof withstanding 10 psi internal pressure.

Laser Welding. A CO2 laser was used success-fully to thermally weld the two halves of the plasticstructure. Figure 2 shows devices that were bondedtogether using this technique. Maintaining intimatecontact between the plastic sheets was the primarychallenge for laser welding these structures. For flatsheets, we used a vacuum platform to suction thework together and cut the top sheet larger than thebottom sheet. For embossed sections, we madenesting fixtures out of acrylic. We adjusted theheight and spread of the beam to get reasonablewelds. We were able to produce welds which held tointernal pressures of 15 psi. However, we were notable to get within 0.02 in. of the edge of the raisedembossed plastic because the two plastic filmscould not be reliably held together near that inter-face. Also, we could only weld a flat sheet with oneembossed side, or two sheets that were embossedsimultaneously and in the same direction. Laterwelds performed for a PCR sample preparationsystem were not leak-free due to the difficulty ofcontrolling the very low energy density level of theavailable laser.

Silicone Molding. We turned to silicone for thisapplication because of the increased manufactura-bility of the structures and the higher probability forobtaining good valve shut-off. We used etched siliconwafers to form molds into which we poured a twopart, heat-cured silicone. We used both Sylgard 184and Sylgard 186. We used two mirror-imaged molds


Air

MixerMixer

Output

Actuator

Valve

Valve

Pump

Pump

Plenums

Tubes

Input

Input

Figure 1. Formedplastic structure withexternal actuation.

Figure 2. Embossed and laser-welded plastic sheet.

205 Miles_qk 7/22/99 1:33 PM Page 2-2


to form two halves of the final structure. We curedthese halves, removed them from the mold, thenpainted on an additional amount of the silicone toact as a glue between the two halves. Thus, theentire mold is made from just one material.

The Sylgard 186 is tougher, more tear-resistant,and has better adhesion than the Sylgard 184, butis much more viscous and sticks to the mold toowell and is, therefore, difficult to use. However, wewere able to make devices that could withstand a35-psi internal pressure before delamination of thelayers occurred.

Interconnects. Interconnecting the device to theoutside world proved difficult. For the flat sheets, wedesigned a flared section that could be mated to astandard leur fitting. However, these leaked a littlewhen used directly and needed sealant to hold. Forthe silicone structures, we used the silicone as thesealant for tubes that were inserted into the moldeddevice. The best results were obtained using PEEKtubing, and acceptable results with silicone tubing.Tygon tubing tended to harden up and tear themolded device. Good surface preparation of thetubing was required for a good seal.

Pneumatic Actuation

Pneumatic actuation was used to provide thevalving, pumping, and mixing functions of the fluidiccircuit. One method of providing pneumatic actua-tion was to use silicone-molded buttons that wereheld in an aluminum clamped fixture, which createda silicone membrane that could be distended usingair driven through standard pneumatic fittings. Thesilicone rubber was very flexible, and this methodworked well, particularly for pumping.

The smaller buttons used for valving required ahigher pressure. We found that complete shut-off of

the flow required a channel depth on the order of100 mm for the silicone molded devices, and zerogap for the welded plastic sheet. A second methodfor providing pneumatic actuation is to use off-the-shelf pneumatic cylinders attached to an aluminumplate for providing force against the plastic struc-ture. Pumping and mixing can be accomplishedusing pneumatic cylinders. Additional force forvalving was obtained by replacing the aluminumplate at the end of the pneumatic cylinder with a0.010-in.-wide shim.

We found that we could also spin Sylgard 184over silicon nitride windows formed by etching a sili-con wafer to make silicone membrane actuatorsthat may be more compatible with micromachineddevices. The silicon nitride could be removed by apost-cure dry etch.

System Results

We were able to demonstrate the functions ofpumping, valving, and mixing using these devices.Pumping is shown in Fig. 3. The fluid is forcedfrom the silicone chamber using an externalpneumatic plunger.

Future Work

We are using this technology in two other applica-tions, a sample preparation system for theautonomous point detector program, and a hand-heldPCR system. The heat-staked metallocene structuresare the most cost-effective structures and will serveas the basis for the applications of this technology.We are working to replace the pneumatic actuatorswith low-power piezoelectric actuators and manuallydriven cam devices to make this device more attrac-tive for portable applications.

FY 98 2-3

(a) (b) Figure 3. Pumping actua-tion on silicone chamberusing pneumatic cylinderactuator.

205 Miles_qk 7/22/99 1:33 PM Page 2-3

ub-Micron Lithography with a 5X Stepper


Introduction

We continue to get requests for lithography withfeature sizes <2 µm. From time to time, there areeven requests for sub-micron images. The existingremote wireless sensors and adaptive opticsprojects are designing devices based on 2-µm rules,while the giant magneto-resistive (GMR) sensorproject uses sub-micron features. In addition, we getrequests for Fresnel zone plates and pinhole arrayswith sub-micron feature sizes.

The contact printers in LLNL’s MTC can be useddown to 2 µm, with increasing difficulty because ofdiffraction effects. A new state-of-the-art projectionlithography system with sub-micron printing capa-bility can cost more than $5M. Fortunately, there isan industry dealing in used, refurbished equipmentwith adequate capability. We accepted delivery of aGCA DSW Model 8500 in January 1998. It is nowoperational in the MTC. The system is ten years old,cost $255K, and has been refurbished to perform toits original specifications, which allows the printingof 0.7-µm features.

Progress

We spent several months working with the vendorto install and learn to operate this system. It is a 5Xreduction system, which means the photomasksneed to be designed at five times the size of thefeatures to be printed on the silicon wafer. Thereduction lens is referred to as a T1635P, whichmeans it has a 16-mm circular field of view with anumerical aperture of 0.35. This allows the printingof 0.7-µm features in 1-µm-thick resist, anywherewithin the 16-mm field of view. The illumination

source is a 1000-W mercury lamp which providesi-line illumination (365 nm). Typical exposure timesfor 1-µm-thick resist is 0.5 s. The system has itsown environmental control chamber and atmos-pheric control system (ACS) to regulate tempera-ture, pressure, and humidity.

This system has many automated features thatare used to optimize its performance. For example:

1. RMS—Reticle Management System, used toselect from 10 reticle stores in a rack. Eachreticle is labeled with a bar code.

2. AWH—Automatic Wafer Handler, automati-cally loads and unloads 100-mm-diameter sili-con wafers from cassette holders.

3. AFII—Autofocus System, automatically main-tains best focus as the wafer is stepped fromdie to die.

4. AWA—Automatic Wafer Alignment, does theinitial global alignment of the reticle to thewafer.

5. DFAS—Dark Field Alignment System, used forthe final die-by-die alignment of the reticle tothe wafer. Initial alignment is accomplished bya laser interferometer-controlled x-y stage.Final alignment is better than ±0.15 µm.

To properly use the automatic sub-systems, themain system must first be set up to be within their“capture” range. This is done during initialization ofthe system. The initialization parameters are storedin the MODE program for system operation.

Also, special targets need to be placed on thereticles for these automatic sub-systems to operate.For example, the AWA and DFAS sub-systems bothneed special targets to be placed on the reticles, inthe proper position.

FY 98 2-5

This year we installed a lithography system capable of printing sub-micron features onto siliconwafers. The system is a refurbished GCA DSW, Model 8500, which is an i-line system (365 nm), capableof printing 0.7-µm features anywhere within a 16-mm field of view. This system is now operational inthe Microtechnology Center (MTC) at the Lawrence Livermore National Laboratory (LLNL).

Dino R. Ciarlo and Benjamin P. LawElectronics Engineering Technologies DivisionElectronics Engineering

210 Ciarlo_qk 7/22/99 3:41 PM Page 2-5


Figures 1 and 2 show SEMs of resist imagesprinted in 1-µm-thick JSR 500 positive resist. Thesmallest pedestals in Fig. 1 have a diameter of0.5 µm. In Fig. 2 are pedestals with lateral featuresof 1.4 µm by 1.4 µm. The sidewalls are nearly vertical.

Future Work

This 5X reduction system is now being used forseveral LLNL projects. We continue to learn howto operate the system and are becoming morecomfortable with its capabilities. The system isoptimized to use 100-mm-diameter silicon wafers

that are 500 µm thick. We recently had a projectthat is using 100-mm-diameter wafers only375 µm thick. This required a rather difficultmanual adjustment of the optical column to focusonto the thinner wafers. Most likely, we will begetting other requests to print on “non-standard”substrates. The computer that controls the systemis rather dated and is not very user-friendly, butwe are learning.

Finally, there are now only three members ofthe MTC trained to operate this system, but weexpect to add two to three more to this list ofoperators in the future.


Figure 1. SEM image of features printed in 1-µm-thick positiveresist. The small pedestals have a diameter of 0.5 µm.

Figure 2. SEM image of a square resist pedestal. Resist thick-ness is 1 µm. Area of pedestal is 1.4 µm by 1.4 µm.

210 Ciarlo_qk 7/22/99 3:41 PM Page 2-6

dvanced Packaging for Wireless Microsensor Modules


Introduction

A wireless microsensor module consists of fourbasic parts:

1. sensor—could be a low-g (µg) accelerometer.The sensor makes use of both bulk and surfacemicromachining.

2. interface electronics—to convert the signalfrom the sensor to a usable voltage. The inter-face electronics drive an A/D converter, whichin turn interfaces to the modem.

3. modem electronics—the electronics that takecare of some portion of the protocol, as well asproducing the spread spectrum codes to drive(transmit data) or decode (receive data) theRF portion of the module.

4. RF section—the mixers, oscillators, amplifiers,and antenna switch. This section is the inter-face between the modem and the antenna.

To integrate these dissimilarly processedparts into a compact form factor, it is necessaryto develop advanced packaging processes. Wehave chosen a microaccelerometer sensor todemonstrate a wireless module and establish theintegration technologies.

The next generation of defense and intelligenceactivities requires covert large-area sensingnetworks for extended periods of time (one month toone year). Current micro-electromechanical systems

(MEMS) technologies can enable only a point detec-tion device without communications between multiplenodes. Wireless technologies are not self-sufficientwithout physical transduction capabilities throughMEMS. The combination of the two is of limited useif it isn’t covert, or close to covert. Although thereare many microsensors being developed commer-cially and in academia, none are integrable withwireless communications platforms to enablenetwork sensing.

The challenge lies in the integration of differentchips in a compact form, which requires techniquesto ensure compatibility of process and signal inter-faces. The engineering capabilities supporting theMicrotechnology Center (MTC) at LawrenceLivermore National Laboratory (LLNL) have inter-disciplinary expertise in 1) MEMS-based microsen-sor design; 2) MEMS electronics packaging;3) MEMS electronics sensing design, deep reactiveion etching (DRIE) capability (a key to integratingmicromachining technologies in a compact form);and 4) programmatic applications requiring asystems approach and allowing the generation of astandard protocol.

Integration of the numerous parts of themodule will be designed from the systems level toensure miniaturization, compatibility in elec-tronic interconnect and compatibility in differentprocessing sequences.

FY 98 2-7

We have developed packaging technologies to integrate the components of wireless sensingmodules such as MEMS microsensors, sensor electronics, modems, RF transceivers, and RF elec-tronics, all of which may ultimately be on individual dies.

Abraham P. Lee, Charles F. McConaghy, Jimmy C. Trevino, and Leslie M. JonesElectronics Engineering Technologies DivisionElectronics Engineering

Jonathan SimonNew Technologies Engineering DivisionMechanical Engineering

215 Lee_qk 7/22/99 4:46 PM Page 2-7


Most other researchers are working on singularaspects of the problem, such as the wireless andMEMS technologies. Challenges to combine the stateof the art in MEMS and wireless communications is aunique opportunity, bringing together expertise inelectronics and mechanical engineering, chemistryand materials science, and nonproliferation control.

The technology developed by this project willdirectly enhance core competencies in the MTCsince it will enable the integration of micromachin-ing technologies for many applications.

While there are many researchers in the field,LLNL can distinguish itself as the leader in technol-ogy integration with advanced packaging of wireless

components, large quantity sensor networks, andadvanced MEMS electronics packaging. It alsoallows the opportunity for LLNL to set the stan-dards and protocols for wireless microsensormodules and networks.

For the remote sensing needs that are increas-ingly important in this post-cold-war era, thisproject could have a significant effect on monitoringtreaties and combatting urban warfare. It affects thecapability to maintain world peace and nationalsecurity. The State of California could benefit fromborder control technologies developed from thisproject. This project is in direct alignment withLLNL’s mission of global security.


Glass or alumina carrier

MEMS chip IC chip (Si or other material)

Indium solder bump

Released MEMS structure

Interconnects

Feedthrough

Hermetic seal

Bonded glass cover

Carrier with through-wafer vias

(a)

(b)

(c)

Figure 1. MEMS-integrated MCM packaging scheme.

215 Lee_qk 7/22/99 4:46 PM Page 2-8


Progress

Integration of Wireless Microsensor Module

We have developed packaging technologies tointegrate the components of wireless sensingmodules such as MEMS microsensors, sensor elec-tronics, modems, RF transceivers, and RF electronics,all of which may ultimately be on individual dies.

There are several important constraints to thedesign:

1. Chips may be fabricated from different materi-als and processes, such as Si and GaAs.

2. Chips may be of largely varying size from2 mm × 2 mm, to 1 cm × 1 cm.

3. MEMS sensors may require vacuum, fluid, orgas environments.

4. Post-processing of the chips should be minimal.5. The approach is flexible.We have opted for a Flip-Chip Multi-Chip Module

(MCM) approach. In such a design, the individualchips are flipped over and bonded to a carrier,rather than interconnected using wire bonds. Thehybrid package can then be further encapsulated orrepackaged as necessary.

Figure 1a shows a glass (fused silica or Corning7740) or alumina carrier with patterned intercon-nects and electro-deposited indium solder bumps.All the processing is done on the carrier wafer toensure compatibility and minimize the handling ofthe various dies.

A variety of chips can then be flip-chip-bonded tothe carrier. In this case, a recess for the MEMSdevice has been milled into the carrier. This recesscan serve as an attachment point for fluid or gasinterconnects, or as in Fig. 1b, the indium may alsobe used to seal the MEMS die into a vacuum reser-voir or other filled reservoir.

In the case of extremely space-constrainedsystems, addition of through-wafer vias (Fig. 1c) tothe carrier effectively doubles the packaging density.

A fabrication process flow for a carrier waferwith no vias is shown in Fig. 2. A carrier wafer ofglass (fused silica or Corning 7740) is coated with ametal seed layer of 500 Å Ni/2000 Å Au/100 Å Ti.The gold acts as the primary conductor, while thetitanium acts as an adhesion layer, and the nickelacts as a barrier to keep the plated indium fromdiffusing into the gold.

FY 98 2-9

Metal ion mill

Ni/Au/Ti(a)

Deposit Ni/Au/Ti seed layer

Glass carrier

Polyimide solder dam

Deposit and pattern Polyimide

In Thick photoresist

Pattern thick PR mold and plate Indium

Reflow

(b)

(c)

(d)

(e)

Figure 2. Process flow for carrier wafer traces and indiumsolder bumps.

215 Lee_qk 7/22/99 4:46 PM Page 2-9


In the first mask step, a polyimide layer ispatterned which will be used both as a solder damfor the indium and to define the traces. In the secondmask step, a thick photoresist is spun onto the waferand patterned to make a mold for indium plating.

After indium plating (~50 µm), the photoresist isstripped, and the parts are ion-milled to pattern thetraces. The indium is then reflowed in an inert envi-ronment to form uniform bumps. Figure 3 shows35-µm-tall indium bumps before reflow.


Figure 3. SEM of ~35-mm-tall electroplated indium bumpsbefore reflow.

215 Lee_qk 7/22/99 4:46 PM Page 2-10

adio Frequency (RF) Technology for Wireless Microsensor Modules


Introduction

Networks of low-power wireless sensors are envi-sioned to accommodate future sensor needs in thefield. Such packages should not only includesensors, such as a seismic or bio-sensor, but alsothe means to transmit this information to somedistant monitoring point. These sensors should bevery low power and small in both size and weight toallow for ease in deployment. The communicationdistance of any module is about 30 to 100 m, andthe RF power is 1 mW. However, networks of thesemodules can be assembled to cover much longerdistances. The module being constructed this year isa cube 2 in. on a side. The RF components occupyabout one third of the volume, the batteries andsensor occupy the remaining volume.

Progress

The RF section consists of a modem which isused to both modulate and demodulate the digitaldata onto and from a 900-MHz RF carrier that hasbeen implemented with a field-programmable gatearray (FPGA). In addition, on the sensor end of theRF link, the modem is also used to read the A/Dconverter that digitizes the sensor output. An addi-tional operation of the FPGA is the superposition ofa pseudo-random code on the data at transmit, andthe removal of this code upon reception. Besides the

modem, the necessary RF circuits, which either up-convert on transmission or down-convert on recep-tion, the modem signals have been implemented ona pc board that is 2 in. square. These circuits weredesigned and built during FY-98.

A time division protocol has also been imple-mented which allows each sensor module tocompletely power its RF circuitry off except whenneeded. In low-power applications, even the receivercan cause a significant battery drain. The time divi-sion protocol allows each sensor module to powerup its RF circuitry for 260 ms and then be off forabout 2 s to save power. There is a sophisticatedresynchronization process that occurs periodicallyto keep the clocks in the sensor module in sync withthe base station that is interrogating the modules.

Future Work

Future work requires that we make even smallerand lower power RF circuitry, such that smallerbatteries can be used for longer periods of deploy-ment. In addition, the overall sensor module needsto fit a package that is about one cubic inch involume, which means that the RF circuitry needs tobe decreased in size. This will be achieved by usingapplications-specific integrated circuits (ASICs)rather than FPGAs, and RF chips with more func-tionality per unit of volume.

FY 98 2-11

We have developed a small, low-power RF link for connecting sensors to a central node. The link isin the 900-MHz ISM band and uses spread spectrum technology, as well as time division multipleaccess, to accommodate multiple sensor modules. This was a joint project with researchers at theUniversity of California, Los Angeles (UCLA). The modem and RF technology developed at UCLA isbeing incorporated into a Lawrence Livermore National Laboratory wireless sensor package.

Charles F. McConaghy and Abraham P. LeeElectronics Engineering Technologies DivisionElectronics Engineering

Charles Chien, Chris Deng, and Igor ElgorriaUniversity of California, Los Angeles

220 McConaghy_qk 7/22/99 4:52 PM Page 2-11

ltra-High-Speed Analog-to-Digital Conversion Technology


Introduction

Many national security missions, particularlythose that are intelligence community-related,involve the application of advanced communication,instrumentation, radar, sensor, and electronicwarfare systems. These all rely on ADCs to digitize alarge information bandwidth (GHz) with highdynamic range and precision. However, the perfor-mance of state-of-the-art ADCs has progressedrather slowly—about 1-bit improvement, or doublingin sampling speed every six to eight years. This can,to a large extent, be attributed to the limitations ofthe available semiconductor technologies in terms ofdevice matching, device operating frequencies, andnoise and nonlinearity in active devices.

To obtain a quantum leap in performance beyondthat of current electronic ADCs, we propose todevelop the enabling technology for a class ofphotonic ADC architectures based on advancedoptoelectronic technology. With the unique ultra-high frequency capability of advanced optoelectronic

components, the proposed class of photonic ADCswill simultaneously attain high sampling rates andlarge dynamic ranges. These photonic ADCs, alongwith advanced sensor technology, will allowmeasurement of physical phenomena of nearly everytype with unmatched speed and accuracy. For appli-cations that require high precision, but not neces-sarily fast sample rates, photonic ADCs will enableoversampling at unmatched sample rates to enableultra-high precision sigma-delta ADC architectureswhich trade-off sample rate for precision.

Progress

Our conceived ADC photonic architecture (Fig. 1)uses a multi-wavelength ultra-short laser pulse trainand a Mach-Zehnder modulator to sample a broad-band signal of interest (Fig. 1, RF signal in). Thesampled signal is temporally demultiplexed througha wavelength division multiplexer to an array ofphotodetectors where the outputs can then be digi-tized by a time interleaved array of slower-speed

FY 98 2-13

We have evaluated competing photonic approaches to the problem of high-speed analog-to-digitalconversion (ADC). Our study of the existing literature has led to a novel technology that will enableextremely high-fidelity ADCs for a variety of national security missions. High-speed(>10 Gigasamples/s), high-precision (>10 bits) ADC technology requires extremely short aperturetimes (~1 ps) with very low jitter requirements (<10 fs). These fundamental requirements, along withother technological barriers, are difficult to realize with electronics.

We have conceived a multi-wavelength approach that uses a novel optoelectronic soliton. Ourapproach uses an optoelectronic feedback scheme with high optical Q to produce an optical pulsetrain with ultra-low jitter (<5 fs) and high amplitude stability (<10–10). This approach requires lowpower and can be integrated into an optoelectronic integrated circuit to minimize the size.

Mark E. LowryPhysics and Space Technology

Ronald E. HaighDefense Sciences Engineering DivisionElectronics Engineering

Charles F. McConaghy Electronics Engineering Technologies DivisionElectronics Engineering

230 Lowry_qk 7/22/99 4:55 PM Page 2-13


electronic ADCs. We have also envisaged morecomplex photonic ADC architectures. All require theproposed optical source to sample a signal of interest.

The multi-wavelength comb laser is realized witha novel externally modulated coupled-cavity Fabry-Perot interferometer with optoelectronic feedback(Fig. 2). Here the RF input driving the phase modu-lator will periodically sweep the optical pass-bandthrough the cw wavelengths, simultaneously pulsemodulating each wavelength and providing theappropriate temporal spacing for each wavelengthin the resulting comb.

A portion of the optical output signal is routedthrough a fiber delay line back to the phase modula-tor input, thus, creating a high Q optical feedbackcircuit. The signal is then bandpass-filtered opticallyto select only one wavelength of the multi-spectralpulses for detection. The detected signal is band-pass-filtered electrically to eliminate harmonics, andamplified with a low-noise narrowband RF amplifierprior to driving the phase modulator.

The laser source will generate an optical pulsetrain with pulse widths between 1 and 10 ps and ajitter of 1 to 10 fs. We expect to realize a low-jitteroptical pulse train with less than 5 fs of jitter over atleast a 1-ms integration window using thisapproach. This jitter specification is at least anorder magnitude improvement over state-of-the-artmode-locked feedback loop semiconductor and fiberlasers. These lasers have relatively large pulse jitterbecause of the amplified spontaneous emission(ASE) noise present in the lasers.

Because we modulate external to the gainmedium, and in the absence of ASE, the timing jitterof our source will be governed exclusively by theeffective Q of the optoelectronic circuit and the noise

of the RF amplifier driving the external phase modu-lator. The long term stability will be governed by themechanical and temperature stability of the electro-optic Fabry-Perot cavity and the fiber feedback loop.

The multi-wavelength nature of the optical pulsetrain is required to temporally demultiplex the opti-cally sampled signal. The Fabry-Perot cavity can besimultaneously seeded with lasers of differentwavelengths resulting in a multi-wavelength pulsetrain. Our approach synthesizes the generation ofthe required ultra-short pulses from an array ofcontinuous wave lasers. Optical pulses withpulsewidths as short as 660 fs have been demon-strated using electro-optic synthesis.

Recent Accomplishments

Our accomplishments in FY-98 included the following:

1. Several classes of photonic ADC architectureshave been conceived. These architectures havemany common features including the notion ofsampling a signal with a multi-wavelengthoptical pulse train to allow the signal to bedemultiplexed spectrally.

2. A novel multi-wavelength optoelectronicsampling source has been envisioned. Thesource develops ultra-short optical pulsesusing external phase modulation in a Fabry-Perot cavity. This approach enables opticalpulse trains to be generated with ultra-lowjitter characteristics along with high amplitudestability.

3. A mid-year proposal was funded, based onthese ideas. Some experimental resultshave been obtained.


Electrical outRF signal in

Opticalmodulator

δt

λ1 λ2…λΝ

λ2

λN

λ1

λ1 λ2…λΝ λ1 λ2…

T

T

T

… …WDMdemux

Mul

ti-λ

co

mb

lase

r

∆t

T

…

Figure 1. OpticalSampler/demultiplexerphotonic front-endbuilding block.

230 Lowry_qk 7/22/99 4:55 PM Page 2-14


4. A proposal to continue the development of this work with an external sponsor hasbeen developed.

Finally, other applications for the proposedsource have been considered, including 1) synthesisof optical sidebands with precise frequency separa-tion for optical spectroscopy; and 2) wavelengthencoding schemes for ultra-high (100 fs) temporalresolution of transient events.

Future Work

It is anticipated that this effort has a highprobability of securing funding in FY-99 and issynergistic with other high-speed instrumenta-tion projects at Lawrence Livermore NationalLaboratory including FemtoScope, RadSensor,and Photonic Doppler Velocimetry.

FY 98 2-15

RFamplifier

Fiberdelay

line

Multi-λ comb laser

N λ

… … …MUX

Bandpass filter

RedBPF

Multi-λ CWlaser

MM M

Phase modulator

Figure 2. Coupledcavity multi-wave-length comb sourcewith high-Q opto-electronic feedback.

230 Lowry_qk 7/22/99 4:55 PM Page 2-15

igh-Voltage Photovoltaics


Progress

This year, we undertook a development effort tosuccessfully demonstrate kilovolt potentials in anelectrically isolated manner using photovoltaics anda fiber optic laser source.

In FY-97, our research and development led us toinvestigate approaches to reducing the photonsabsorbed in the semi-insulating substrate betweenadjacent cells. These photons induce photoconduc-tive current between the devices, which tends toshunt the cells. By depositing an absorptive orreflective material on the semi-insulating substrateregions of the device, we could successfully shieldthe substrate from photons and eliminate the photo-conductive current in the substrate.

We designed prototype photovoltaic linear arrays,including the epitaxial structure of the devices,tested the functional, monolithically-integrated 90-Vphotocells, and evaluated techniques to stack thesephotocells to generate multiple kilovolts of potentialin a compact area. We successfully demonstrated a1260-V photocell by dicing the individual 90-Vphotocells and wirebonding fourteen of them inseries on a glass substrate.

The photocell is shown in Fig. 1.

FY 98 2-17

This technology-base project is a continuation of research to develop photovoltaic technology togenerate voltages up to several kilovolts for powering electrically isolated systems with a fiber opticlaser source.

Karla G. Hagans and Ronald E. Haigh Defense Sciences Engineering DivisionElectronics Engineering

Figure 1. Photocell. We demonstrated 1260 V from thisphotovoltaic device.

235 Hagans_qk 7/22/99 4:57 PM Page 2-17

attice Boltzmann Simulation of Complex Fluid Flows


Introduction

Fluid flow has been a topic of engineering impor-tance and research for many years. Traditionalmethods to solve such problems include exacttheory, finite element methods, finite differencemethods, finite volume methods and singularitymethods. All of these approaches have their roots inthe continuum approximation of fluid mechanics orthe Navier-Stokes equations. When applied appropri-ately, these approaches provide a qualitative andquantitative description of flow systems. A relativelyrecent approach, the LB method, has been growingin utility and popularity in the fluids community.

The LB method has its roots in the kinetic theoryof gases1 or the Boltzmann transport equation. Inthe limit of small Mach number and small Knudsennumber, the Boltzmann transport equation recoversthe Navier-Stokes equations of motion. Because theBoltzmann transport equation describes the statis-tical average motion of fluid molecules, this is notsurprising. In contrast, the continuum approachescited above are macroscopic descriptions thatassume that the properties of interest, for example,density, and fluid velocity, are continuous through-out space.

The LB equation can be thought of as the discreteform of the Boltzmann transport equation. Like theBoltzmann transport equation, the LB equation alsorecovers the Navier-Stokes equations of motion inthe limit of small Mach and Knudsen numbers. TheLB method provides great flexibility in the study ofbounded, particulate media, or complex fluids.

Essentially, the study of complex systems usingthis approach is limited by the users’ ability todescribe the medium. For example, a core sample ofsand stone was imaged and mapped onto a lattice,permitting the study of two-phase, oil/water, flowthrough the semi-permeable medium.2 Additionally,the LB method has only recently found a niche in theengineering community, and therefore is still a novelapproach to such problems.

For a detailed description of the theoreticalconsiderations and breadth of application, thereader is encouraged to see the work of Chenand Doolen.2

Progress

Selected results from a recent article3 andcurrent research efforts are presented in this reportto demonstrate the utility of the LB method.Specifically, results for the velocity profile in a Hele-Shaw cell are presented, and the hydraulic perme-ability calculations for the BCC lattice of cylindersas described by Higdon and Ford4 using the LBmethod are compared with their Spectral BoundaryElement results for the same system. In the conclu-sions a brief description of future work at LawrenceLivermore National Laboratory (LLNL) is discussed.

The Hele-Shaw Cell

Many flow systems of interest involve boundedflows. A well-known example of such a system is theHele-Shaw flow cell. The Hele-Shaw cell was put

FY 98 2-19

This report describes the lattice Boltzmann (LB) method for simulation of complex fluids. LB simu-lation results for a few flow configurations are considered; namely, results for flow between infiniteparallel plates, the Hele-Shaw cell, and flow through a 3-D, ordered array of cylinders. The ability ofthe LB method to correctly predict the fluid velocity profile in the Hele-Shaw cell is examined, and theerror as a function of lattice site density is discussed. The hydraulic permeability, which is a functionof the hydrodynamic force acting on a stationary object due to fluid flow, for a 3-D, ordered configura-tion of cylinders is compared with the results of Higdon and Ford.

David S. ClagueNew Technologies Engineering DivisionMechanical Engineering

245 Clague_qk 7/22/99 4:59 PM Page 2-19


forth as a representation of a porous medium, asshown in Fig. 1.

The flow cell is periodic in both the x1 and x2directions; hence, the unit cell gets replicatedthroughout space in the x1-x2 plane. The imposedflow is in the x1 direction. Essentially, the verticalgap, H, between bounding walls represents theequivalent, average distance between fixed obsta-cles in a porous medium.

LB simulations were performed to study pressure-driven flow in this configuration and compared withthe analytic solution. In lattice space, a pressuregradient is imposed by applying a constant bodyforce on the fluid lattice sites between the walls. Arepresentative fluid velocity profile is compared withexact theory in Fig. 2. The error between the LBresult and the analytic solution is approximately0.5%, on average.

Similar simulations were run for various nodedensities across the gap, H, and the spatial conver-gence is approximately second order.

Three-Dimensional BCC Lattice of Cylinders

Flow through fibrous media is common in manyfiltration systems. Examples range from man-madeair filters to blood filtration in the kidney; hence, theability to predict flow behavior in such systems is ofparamount importance. The BCC lattice4 provides arigorous test of the LB method for flow through 3-Dfibrous media. An LB representation of a typical BCCsimulation cell is shown in Fig. 3.

The solid lattice sites represent the solid phase,and the clear or unmarked lattice sites represent thefluid phase. For visual clarity, the cylinder radius isonly three lattice sites, but for the actual simulationresults, the cylinder radius was increased toincrease the force resolution at higher fiber volumefractions, φ. Flow is imposed here, again, by apply-ing a uniform body force on the fluid lattice sites. Toenforce the no-slip condition in the LB method, thefluid is “bounced back” to originating lattice sitesthat are adjacent to the solid surfaces.2,3 Therefore,knowing the pressure gradient, ∇ P, the superficialaverage fluid velocity, ⟨u⟩ and the fluid viscosity, µ,one can make use of Darcy’s law,

(1)

and calculate the hydraulic permeability, k, ofthe medium.

In the results shown in Fig. 4, the actual cylinderradii ranged from 5 to 30 lattice sites or lattice units.

uk

P – ,= ∇µ


HU

23x

x

x1

Flui

d v

elo

city

Position (lattice units)

0.04

0.05

0.06

0.03

0.02

Lattice BoltzmannTheory

0.01

0 5 10 15 20 3525 300

60

60 60

40

4040

20

2020

0

00

Figure 1. Hele-Shaw cell. The flow direction is in the x1 direc-tion, and the simulation cell is periodic in both the x1 and x2directions. H is the vertical distance between the solid, infiniteparallel planes.

Figure 2. Fluid velocity profile. The LB result is compared withexact theory for pressure-driven flow in a Hele-Shaw cell. Theseparation, H, in lattice space is 38 lattice sites.

Figure 3. Three-dimensional, BCC lattice. The cylindersdepicted have a radius of three lattice units. Cylindersemanate from the center of the cell and intersect the eightvertices of the cubic domain. The dimensions of the simulationcell are 60 x 60 x 60, which represents a fiber volume fractionthat is approximately equal to 0.36.

245 Clague_qk 7/22/99 4:59 PM Page 2-20


As shown, the hydraulic permeability, madedimensionless with the cylinder radius, r,2 is plottedas a function of fiber volume fraction, φ. Here LBsimulation results are compared with the result fromthe Spectral Boundary Element Methods of Higdonand Ford4 for the BCC lattice configuration. Theirapproach is both rigorous and accurate for nearly allpossible fiber volume fractions. As is clearly seenhere, the LB results are nearly identical to that ofHigdon and Ford4 over the entire range of fibervolume fractions considered. This is remarkablesince many methods fail when considering fibervolume fractions >0.5.

Future Work

The LB method is indeed a powerful tool for thestudy of fluid flow through complex media. The simu-lation results above confirm that it is as accurate asthe best methods available, Spectral BoundaryIntegral Methods, for solving Stokes flow problems.The LB results represent building blocks for model-ing and simulation of systems of greater complexity.Our research focuses on using LB methods to studytransport phenomena issues relevant to current andfuture efforts in LLNL’s Microtechnology Center.

References

1. Landau, L., and E. M. Lifshitz (1989), “A Course ofTheoretical Physics, Volume 10,” Pergamon Press.

2. Chen, S., and G. D. Doolen (1998), “LatticeBoltzmann method for fluid flows,” Ann. Rev. FluidMech., 30, pp. 329-364.

3. Clague, D. S., B. D. Kandhai, R. Zhang, and P. M. A.Sloot (1998), “The hydraulic permeability of (un)bounded fibrous media,” submitted to J. Fluid Mech.

4. Higdon, J. J. L., and G. D. Ford (1996) “Permeabilityof three-dimensional models of fibrous porousmedia,” J. Fluid Mech., 308, p. 341.

FY 98 2-21

Hyd

raul

ic p

erm

eab

ility

(k/

r2 )

Fiber volume fraction (φ)

0.1

1

10

0.01

0.001

Lattice BoltzmannBCC latticeHigdon and Ford (1996)

0.0001

0 0.2 0.4 0.6 0.8 110

-5

Figure 4. Hydraulic permeability for the BCC lattice of cylin-ders. LB results are compared directly with Spectral BoundaryIntegral results.

245 Clague_qk 7/22/99 4:59 PM Page 2-21

icro-Electromechanical Systems (MEMS) forCharacterization of Plastic-Bonded Explosives


Introduction

One of the key problems in support of stockpilestewardship is the characterization of aging mecha-nisms for polymer binder materials used in PBX invarious shapes and housings. Understanding themechanical properties of PBX materials systemsunder accelerated aging conditions can provide valu-able insight into the failure mechanisms of thesematerials, and subsequently, the lifetime of stockpilesystems and devices.

Typical high-explosive (HE) crystallites in bindermaterials have dimensions on the order of 10 to50 µm, with distances between crystallites as smallas 5 µm. While mechanical properties of the bindermaterial characterized at the macro-scale areimportant, it is critical to understand the micro-scale

properties of the binder material, as well as theHE/binder interface.

Micromachined structures and devices providecharacterization tools with these properties over µmscale dimensions. At present, micromachining tech-nologies have focused on the fabrication of custommicro-probes for AFM nano-indentation measure-ments and custom templates to form reservoirs ofbinder materials for calibrated nano-indentationexperiments. Nano-indentation enables the time-dependent mechanical properties of the bindermaterial to be measured. With a fixed load appliedto the AFM-nano-indenter, the micro-scale equiva-lent of a creep test is performed by monitoring howthe nano-indenter plastically displaces with time. Inthis manner, the viscoelastic properties of the bindermaterial can be characterized.

FY 98 2-23

Nano-indentation using custom microprobes within an atomic force microscope (AFM) can be anextremely valuable tool for characterization of mechanical and viscoelastic properties of materials atthe microscale level. We have used AFM nano-indentation to characterize the mechanical propertiesof mock plastic-bonded explosives (PBX). Characterization of the materials properties at the micro-scopic level enables better understanding of the aging mechanisms of these materials.Micromachining techniques have also been used to design calibrated experiments to assist in inter-preting the nano-indentation results under various circumstances.

Jeffrey D. Morse Electronics Engineering Technologies DivisionElectronics Engineering

Dino R. Ciarlo Engineering Research DivisionElectronics Engineering

Scott E. Groves and Diane J. ChinnManufacturing and Materials Engineering DivisionMechanical Engineering

Mehdi BaloochDefense and Nuclear Technologies

Mark J. LaChappelDefense Sciences Engineering DivisionElectronics Engineering

TA250 Morse_qk 7/22/99 5:02 PM Page 2-23


Progress

Nano-indentation measurements were performedon mock PBX using an AFM system at LawrenceLivermore National Laboratory (LLNL). The AFM ismodified with a transducer-indenter assembly calleda Micromechanical Testing Instrument. Like itsconventional counterpart, the instrument can imagethe topography of specimens by tracing the superficialcontours of the sample with a constant sub-µN load.

In addition, the device is a force-generating anddepth-sensing instrument capable of generatingload-displacement curves at specified locations. Thelocal mechanical properties such as stiffness,

hardness, and elastic modulus can be determined.The stiffness is defined as the slope of theforce/displacement curve during unloading. Theelastic modulus, E, and hardness, H, are defined as:

where S is the contact stiffness, Fmax is the maxi-mum load and a is the projected contact area underthe load. The resulting force vs time information isillustrated in Fig. 1.

While these results provide important informationregarding the mechanical properties of moderatelyhomogeneous binder material, the problem becomesmuch more complex with the addition of HE crystal-lites. For this case, custom microprobes of predeter-mined size and shapes must be used, typically withtip shank diameters <0.5 µm and aspect ratios>10:1. These can be designed and fabricated bymicromachining techniques and mounted on theAFM fixture.

It would be desirable to be able to characterizethe mechanical properties of the HE/binder inter-face, as well as regions between crystallites asnarrow as 5 µm. This desire arises due to concernsthat the interface becomes fatigued by aging, leadingto delamination and eventual failure of the material.

E a S H F a / /max=

=π1 2 1 22 and


25050 150100 2000

170

150

110

90

70

50

30

10

130

Displacement (nm)

Forc

e (µ

N)

Figure 1. Force vs time information.

(b) Custom-designed AFM tip.

Very smallradius tip (< 5nm)

(a) Commercial style AFM tips.

90˚-cone angle

Very small radius tip (< 5nm)

Long narrow shank

Figure 2. Sample nano-probe showing that (a) existing AFMtips cannot be used to probe surfaces with deep and narrow crevasses, and (b) tall thin probes can reach deepinto crevasses.

Figure 3. Structure formed by electroplating through atemplate patterned in a resist film.



To accurately characterize the mechanical proper-ties of the interface and narrow regions, the AFMnano-indenter probe can be scanned across theHE/binder interface. Using this technique, a smallerdiameter (<0.5 µm) probe is required which is cylin-drical, with an aspect ratio on the order of 10:1.Thus, the probe exerts force on a constant surfacearea which can access extremely narrow areas.

An example of the required nano-probe is illus-trated in Fig. 2, which compares this requirement torelatively standard AFM probes. A variety of micro-fabrication techniques can be used to realize thissmall-diameter, high-aspect-ratio structure.

Figure 3 shows a 1-µm-high, 0.1-µm-diameterstructure formed by electroplating through atemplate patterned in a resist film. Similar struc-tures can be formed through precision plasmaetching techniques and are presently under devel-opment in a structure which can be retrofit ontothe AFM system.

Finally, a custom template to form reservoirs wasfabricated using photolithography to define patternsin a silicon nitride mask on a silicon substrate with

areas ranging from 5 to 25 µm on a side. Wells werethen etched in the silicon with various depths rang-ing from 5 to 25 µm. This is shown in Fig. 4. Thewells are then filled with binder material, and AFMnano-indentation measurements are performed asmeans to calibrate the resulting mechanicalresponse of the binder in proximity to the interfaceand as a function of depth to a hard surface beneaththe probe region. This will provide important infor-mation regarding the three-dimensional nature ofthe samples being characterized.

Future Work

Custom reservoir templates have been fabricatedin silicon and are presently being prepared for depo-sition of mock PBX. AFM nano-indentation measure-ments will be performed to provide a comparisonbetween standard AFM probes and custom AFMprobes, as described above. The custom AFMprobes must be fabricated in a structure which isreadily mounted in the AFM system for calibrationand measurements.

FY 98 2-25

(a) (b) Figure 4. Customtemplate micro-machined in siliconsubstrate for calibra-tion studies of nano-indentation tech-niques on mock PBX:(a) photograph; (b)SEM image.


icro-Electromechanical-Systems-(MEMS)-BasedFuel Cell Technology


Introduction

A serious need exists for portable power sourceswith significantly higher power density, longeroperating lifetime, and lower cost. Presentrechargeable and primary portable power sourceshave excessive weight, size and cost, with limitedmission duration. As an example, batteries coveringa power range from 1 to 200 W have specificenergies, ranging from 50 to 250 Wh/kg, whichrepresents 2 to 3 h of operation for a variety ofcommercial and military applications.

An alternative power source is the fuel cell, whichpotentially provides higher performance powersources for portable power applications if the stackstructure, packaging, and cell operation are madecompatible with scaling down of size and weight.

Fuel cells typically consist of electrolyte materials,either polymer or solid oxide, which are sandwichedbetween electrodes. The fuel cell operates bydelivering fuel (usually hydrogen) to one electrode,and oxygen to the other. By heating the electrode-electrolyte structure, the fuel and oxidant diffuse tothe electrode interfaces, where an electrochemicalreaction occurs, releasing free electrons and ionswhich conduct across the electrolyte.

Typical fuel cells are made from bulk electrode-electrolyte materials which are stacked and mani-folded using stainless steel packaging. Thesesystems are bulky and operate at high temperatures(>600 °C). If the electrode-electrolyte stack can be

made very thin and deposited using thin-film deposi-tion techniques, the temperature of operation will besignificantly lower.

Previous efforts at Lawrence Livermore NationalLaboratory (LLNL) have demonstrated the synthesisof a thin-film solid-oxide-based electrolyte fuel cell(TFSOFC) stack.1,2 The TFSOFC stack was formedusing physical vapor deposition (PVD) techniques.The host substrate was a silicon wafer covered by athin layer of silicon nitride. A layer of nickel was firstdeposited, followed by a layer of yttria-stabilizedzirconia (YSZ).

The conditions during the deposition wereadjusted to achieve smooth, dense, continuous films,thus avoiding pinhole formation which could resultin electrical shorting through the electrolyte layer.This enables the electrolyte layer to be on the orderof 1 µm thick, rather than typical thicknesses on theorder of >10 µm for bulk solid-oxide fuel cells.

By thinning the electrolyte layer, resistive lossesare significantly lower, and the fuel cell operates atmuch lower temperatures. A silver electrode layer isdeposited on top of the YSZ layer. The depositionconditions of this film are adjusted to create aporous structure so that oxygen can readily diffuseto the electrolyte interface.

Progress

During the past year, this effort has focused onthe utility of a TFSOFC, along with assessments for

FY 98 2-27

We have fabricated a fuel cell stack which uses thin-film electrodes, catalysts, and ion-conductinglayers deposited by physical vapor deposition techniques. The stack has been patterned with electri-cal connections using standard micro-fabrication techniques, and subsequently formed into free-standing membranes by micro-machining away the silicon substrate. Manifold structures have alsobeen fabricated through silicon micro-machining techniques.

Jeffrey D. Morse and Robert T. GraffElectronics Engineering Technologies DivisionElectronics Engineering

Alan F. Jankowski and Jeffrey P. HayesMaterials Science and Technology DivisionChemistry and Materials Science

TA255 Morse_fuel_qk 7/22/99 5:08 PM Page 2-27


other electrolyte materials systems. The incorpora-tion of manifold structures within the host substratethrough micro-machining techniques enables acomplete fuel cell device to be realized which can bereadily attached to fuel and oxidant sources. Thisconcept is illustrated in Fig. 1.

In this approach, the fuel cell stack is created bythin-film deposition techniques. Integrated-circuittype micro-fabrication processes are used to patternelectrode contacts, as well as to form a resistiveheater element within the stack structure.

The stack is subsequently formed into a free-standing membrane by selective etching of thesubstrate. Manifold channels are micro-machined inanother substrate, which is subsequently bonded tothe fuel cell substrate, as illustrated in Fig. 1.

This approach provides an effective means toform efficient fuel cell stack and electrode structuresmonolithically, and distribute fuel to the entire stackwithout the need for bulky complex manifolding.Furthermore, since the stack is now only a smallpercentage of the mass of the entire structure,

appropriate thermal design of the fuel cell device,package, and resistive heating elements will allowefficient, low-power heating of the stack.

Figure 2a illustrates a completely fabricatedfuel cell module with integrated heating element,with a view of the fuel cell stack free-standingmembranes shown in Fig. 2b. To achieve this, thesilicon substrate was selectively etched with potas-sium hydroxide, using patterned silicon nitride asthe mask.

Manifold channels, shown in Fig. 3, were etched ina silicon substrate using similar techniques. Thesecomponents will ultimately be bonded together to forma fuel cell module having inlet and outlet channelswith approximately 50-µm-×-200-µm openings forfuel delivery.

The MEMS-based thin-film fuel cells are tested tomeasure the current output as a function of temper-ature as the voltage is incremented from nil throughand above the open circuit voltage (OCV). The solid-oxide fuel-cell (SOFC)-layered combination of a Ni-YSZanode, YSZ electrolyte, and Ag-YSZ cathode, allows


Electrode

Electrolyte

Micromachinedmanifold system

Membrane-electrodeAssembly (MEA)Present electrode/electrolyte material:

Ni/Yttria stabilized zirconia (YSZ)/Ag

Resistive heater

Electrode

Heater isolation

Figure 1. MEMS-based fuel cell concept using silicon micro-machined host structure and manifold system with fuel cell stack membranes.



for the transport of oxygen ions through the elec-trolyte via a diffusion-moderated process. The resis-tivity of the electrolyte is measured to typicallyexceed 1 ΜΩ-cm.

Once an oxygen ion diffuses through the elec-trolyte and combines with hydrogen, an (electron)current is generated. A maximum OCV of ~1.1 Vexits for the combination of oxygen and hydrogenusing this SOFC structure.

The TFSOFC sits atop a windowed silicon waferthat is bounded on the anode side to a quartz tubeusing a ceramic epoxy. The anode and cathode tabson the silicon wafer are silver-epoxy bonded to silverwires. The wires are connected to a semiconductorparameter analyzer that controls the applied

cell potential. The silicon-wafer-mounted tube isO-ring-sealed within a larger diameter quartztube that is placed within a conventional(Lindbergh) clam-shell furnace.

Feedthroughs are provided at either end of theassembly, on both the anode and cathode sides, thatallow the passage of the oxidant and fuel as gases. Afuel mixture of 3% hydrogen is flowed through theanode tube at a rate of 1 to 3 sccm. Air is flowed tothe cathode surface at a rate of 1 to 3 sccm. Thecurrent-voltage output is measured as a function oftemperature to 600 °C.

Initial results of this° testing are illustrated inFig. 4. While not optimal, these results exhibit theexpected overpotential for this electrolyte materialssystem with no output current, along with increasingcurrent output as temperature increases.

While the output current densities are low,inherent limitations are present in fuel cell perfor-mance resulting from the high density of the nickelcathode layer. Thus, while fuel can readily diffusethrough the nickel film to the electrolyte interface,water, the byproduct of the electrochemical reac-tion, is unable to diffuse away from the interface.Thus the reaction ions quench, resulting in limitedefficiency of the fuel cell stack.

Future Work

The next iteration of fuel cell modules willinclude porous nickel electrode structures, therebyeliminating the effects of water vapor trapped at theelectrolyte interface. Further efforts will focus on

FY 98 2-29

Fuel cell module

Electrical contacts forheater control andoutput power

(a) (b)

Figure 2. (a) Micro-fabricated fuel cell module, with (b) free-standing thin-film fuel cell stack membrane, formed by micro-machiningsilicon from backside.

Figure 3. Fuel cell module with micro-machined manifoldchannels in bottom wafer.



optimizing the design of the heating element andfuel cell stack, and packaging. This includes investi-gation of alternate electrolyte material, both solid-oxide and solid-polymer systems.

References

1. Jankowski, A. F., and J. D. Morse (1998), MaterialsResearch Society Symposium Proceedings Vol. 496,pp. 155–158.

2. Jankowski, A. F. (1998), U.S. Patent No. 5,753,385,May 19.


200 °C250 °C300 °C350 °C400 °C450 °C500 °C

1.0

0.8

0.6

0.4

0.2

0.0151050

Current density (mA/cm2)

Cel

l po

ten

tial

(V

)

Figure 4. Current-voltage characteris-tics of thin-film solid-oxide fuel cell atvarious temperatures.


igh-Power GaN Microwave Device Technology


Introduction

GaN is an ideal semiconductor material systemfor high-power microwave applications due to itsunique properties: high thermal conductivity, hightemperature stability, high breakdown field strength,and high carrier velocities. Researchers at theUniversity of California, Santa Barbara recentlydemonstrated1 GaN-based microwave devices oper-ating at cw power densities greater than 3 W/mmgate width, at a frequency of 18 GHz. This perfor-mance is approximately three times higher than thebest commercially available devices made from SiCor InP.

High-power microwave amplifiers presently usedfor communications and radar applications usebulky vacuum tube electron beam devices, such astraveling wave tubes, magnetrons and gyrotrons.Traveling wave tubes, for example, are typically tensof centimeters in size. An all-solid-state, high-powermicrowave amplifier would enable the developmentof much smaller and lighter microwave systems for

use in satellite and ground based communications,and remote sensing. Compact devices and antennasenable the construction of phased-array radars withthe ability to form steerable beams. Moreover, theDC to RF conversion efficiencies for solid-statetransmitters is significantly higher than conven-tional, high-power RF electron beam devices.

Currently, the performance of GaN-based RFdevices is limited by high operating temperatures.This is primarily due to resistive heating in tworegions of the device, namely the resistance betweenthe metal contact and the doped semiconductorcontact layer (that is, the specific contact resis-tance) and the total resistance of the doped semi-conductor layer. By increasing the number of electri-cal carriers in the semiconductor layer, the resistiveheat generation in both of these regions can bereduced. However, one of the main obstacles to thefull realization of the potential of GaN as a materialfor optoelectronic and microwave devices is the diffi-culty in reaching high enough doping levels in thismaterial for fabricating low sheet resistance layers.

FY 98 2-31

The purpose of this work was to develop more efficient doping processes for use in gallium-nitride(GaN)-based power microwave devices. We developed a pulsed laser processing technique aimed atactivating n-type dopants for GaN field effect transistors. The laser processing is performed using aXeCl excimer laser with a 35-ns pulse length at a wavelength of 308 nm. This technology allowsselective area doping of GaN, presently a major technology roadblock in the fabrication of devices inthis material system.

Glenn A. MeyerNew Technologies Engineering DivisionMechanical Engineering

Gregory A. Cooper and Stacy L. LehewElectronics Engineering Technologies DivisionElectronics Engineering

Thomas W. Sigmon and Daniel ToetInformation Science and Technology ProgramLasers

Steven DenBaars and Umesh MishraUniversity of CaliforniaSanta Barbara, California

260 Meyer_qk 7/22/99 5:11 PM Page 2-31


The most common technique used for doping GaNis incorporation of the dopant atoms into the materialduring crystal growth. Magnesium is the mostcommon dopant used to create p-type material inmetal-organic chemical vapor deposition (MOCVD)growth. The ionization energy of the Mg acceptor inGaN is 150 meV. This value is large enough to resultin only a fraction of a percent of the Mg atoms beingelectrically active at room temperature.2 This lowratio of electrical carriers to dopant atoms forcesthe growth to include a significantly higher level ofMg to achieve the desired carrier concentrations.Incorporation of high dopant concentrationscauses the normally transparent GaN to becomecloudy, indicating large concentrations of crystaldefects. Presently, the p-type carrier concentra-tions achievable with magnesium are in the rangeof 2 to 3 × 1018 holes/cm3, a value that is not highenough to achieve low sheet resistance layers orelectrical contacts.

Ion implantation may offer a practical solution tothe problems associated with achieving high p-typecarrier concentrations in GaN. This technique iswidely used in semiconductor device fabricationtechnology. Several researchers have investigatedthis technique for GaN but have encountered prob-lems that do not appear for Si or GaAs. One of theseproblems is the high temperature required to incor-porate the implanted dopant atoms into the crystallattice. The effective incorporation of the dopantsonto the correct crystal sites and removal of theimplant damage requires annealing temperatures inexcess of 1200 °C. Nitrogen, being a fundamentalconstituent of GaN, has a very high vapor pressureat the required annealing temperatures (>1000 bar).Therefore, annealing GaN material at these tempera-tures requires very robust capping layers to preventthe nitrogen from escaping.

Attempts to use rapid thermal annealing (RTA)have also resulted in thermal budgets well in excessof the levels tolerable for device fabrication. Inaddition, heterojunctions and quantum wells areparticularly sensitive to annealing temperatures inexcess of 900 °C. The time-temperature productmust be limited to prevent serious degradation ofthe device layers.

A promising technique useful for reducing thetime-temperature product is pulse laser annealing.Near-surface heating by absorption of a pulsed laserbeam is expected to allow higher processing temper-atures. This typically is a very fast, non-equilibriumprocess. For example, a 90 nm a-Si film on a low-temperature glass substrate can, following theabsorption of a short laser pulse, fully melt and

recrystallize within 150 ns, while the glass remainsunaltered. A pulsed excimer laser can deposit largeamounts of energy into a thin layer at the GaNsurface in a very short time. Deeper layers, contain-ing sensitive quantum structures, do not experiencethe high temperatures.

The objective of our project was to investigate apulsed laser processing method for improving thedoping activation process for n-type dopants inGaN field effect transistors. The laser annealingactually results in the melting of the implantedlayer. During the recrystallization of the moltenlayer, the doping impurity assumes the appropriatelattice position, allowing realization of the desiredlow sheet resistance contact layer. This processtechnology has been demonstrated at LawrenceLivermore National Laboratory (LLNL) to providevery low electrical resistance contacts on Si, SiGe,and SiC semiconductor materials.

Progress

The initial phase of our research focused on theeffects of the pulsed laser process on intrinsic GaN.Once the laser material interactions were under-stood, laser activation of ion implanted n-type andp-type dopants was investigated. The band gap ofGaN semiconductor materials is approximately3.4 eV, making it ideal for strong absorption of our4 eV XeCl laser (λ = 308 nm) pulse. The laserabsorption depth in the GaN is approximately120 nm, using an absorption coefficient of8.37 × 104/cm at λ = 308 nm.3

The GaN films used in the experiments were2 µm thick, grown in a modified two-flow horizontalreactor on c-plane sapphire, using low-pressureMOCVD. The chemical precursors used weretrimethylgallium (TMGa) and ammonia (NH3). Thesamples were then ion-implanted at an energy of150 keV with Si and Mg to doses of 8 × 1016 cm-2

and 5 × 1014 cm-2, respectively.Laser processing of GaN films is performed using

a XeCl excimer laser (from Lambda Physik,Germany) with a 35-ns pulse length at a wavelengthof 308 nm. The transformation induced in the GaNby this pulse is monitored in situ and in real time bymeasuring the time-resolved transmission (TRT) ofan IR laser beam passing through the center of thespot irradiated by the XeCl laser. The samples areheld in a vacuum cell with a quartz front windowthat is transparent to the excimer laser beam. Priorto laser processing, the sample chamber is evacu-ated to 1 × 10-3 Torr, using a Varian sorption pump.The chamber is then backfilled with flowing nitrogen.


260 Meyer_qk 7/22/99 5:11 PM Page 2-32


Figure 1 shows the IR TRT profiles measuredduring the exposure of an undoped film to pulses atdifferent energy fluences. As seen in this figure, theIR transmission drops dramatically upon laser expo-sure at fluences greater than threshold. This drop isassociated with the heating and melting of the filmfollowing the absorption of the excimer laser light.The molten material blocks the IR beam due to itsmetal-like optical properties.

The transmission signals take more than 2 µs toreturn to their pre-pulse levels for the highestannealing fluence, indicating that the annealedmaterial cools down slowly. This slow cooling rate isattributed to the fact that the thermal gradientacross the film is very small, due to the relatively

deep penetration (120 nm) of the 308-nm light inthe GaN. Ultimately, we intend to confine the melt-ing to a shallower region near the surface.Moreover, optical inspection of the films revealsthat for laser exposures above threshold a roughen-ing of the films resulted.

Figure 2 shows the results obtained during theexposure of a sample that was first implanted withMg ions to a dose of 5 × 1014 cm-2, and a depth of100 nm. The most striking feature of this TRT plot isthe fact that the melting duration is now muchlower than that of the unimplanted sample. Thisimplies that the implantation altered the opticalabsorption properties of the film, probably due tosurface amorphization.

FY 98 2-33

0

0.010

0.020

0.030

0.040

-500Time (ns)

IR t

ran

smis

sio

n (

a.u.

)

Laser pulse

0 1000 2000

Above threshold

Undoped GaN

Below threshold

Threshold

Figure 1. IR time-resolved transmissionprofiles obtainedduring laser exposureof undoped GaN filmto various excimerlaser fluences. The IRtransmission dropsdramatically uponlaser exposure withthreshold fluenceindicating thatsurface meltingoccurred.

0

0.010

0.020

0.030

0.040

0 500 1000 1500

IR t

ran

smis

sio

n (

a.u.

)

Above threshold

Time (ns)2000

Below threshold

Threshold

Doped GaN

Laser pulse

Figure 2. IR time-resolved transmissionprofiles resultingfrom laser exposureof Mg ion-implantedGaN film (Mg dose =5 x 1014 cm-2) tolaser pulses withdifferent energyfluences. The IRtransmission dropsdramatically uponlaser exposure withthreshold fluenceindicating thatsurface meltingoccurred.

260 Meyer_qk 7/22/99 5:11 PM Page 2-33



0.20

0.4

0.80.6µm

200

nm

0

500 nm

0

500 nm

Ga balls on GaN surfaceafter laser processing

Uncapped GaNGaN Ga + N2

AIN surface after laser processing

AIN capped GaN

RMS = 8Å

Figure 3. AFM images of GaN surfaces after laser processing. Use of an AlN cap layer prevents GaN decomposition and retains theoriginal surface smoothness.

1022

1021

1020

0 0.1 0.2 0.3 0.4 0.5Distance from surface (µm)

Silic

on

co

nce

ntr

atio

n (

ato

ms/

cm3 )

No anneal

Above threshold

Below threshold

Threshold

Figure 4. Secondary ion massspectroscopy profileof ion-implanted Si inGaN (dose of8 x 1016 cm-2 at 150 keV) laserprocessed at laserfluences above, at,and below threshold.Silicon redistributionis evident at a thresh-old laser fluence.

260 Meyer_qk 7/22/99 5:11 PM Page 2-34


The curves obtained for higher fluence are similarto those observed for the undoped films. The TRTmeasured for the threshold laser pulse has a muchshorter melt duration (~120 ns), that is, a shallowermelt depth. (Heat flow simulations relating meltduration with melt depths are currently being devel-oped for this material.) This corresponds exactly towhat is required for dopant activation. In addition,the surface of the film annealed at threshold is notnoticeably roughened. However, atomic forcemicroscope (AFM) analysis indicates that thesurface of this sample is covered with Ga droplets,suggesting that some surface decomposition hastaken place (Fig. 3).

To prevent decomposition of the GaN during laserprocessing, the surfaces are encapsulated with analuminum nitride (AlN) capping layer. The AlN wasdeposited by ion beam sputter deposition (IBS) usingan AlN target in a 500-eV N2/Ar ion beam. The IR TRTplots and melting thresholds for AlN coated sampleswere similar to those shown in Fig 2. However, amajority of the AlN cap layer was ablated during thelaser exposure. AFM analysis of the surface of theAlN-capped sample after laser processing revealedthat the surface was intact and that no GaN decom-position had occurred.

GaN films ion-implanted with Si to a dose of8 × 1016 cm-2 were coated with a AlN cap layer andthen exposed to laser fluences above, at, and belowthreshold. Secondary ion mass spectroscopy (SIMS)was used to measure the Si profiles of the Si-implanted samples before and after laser processing.

The SIMS data was used to determine the thresholdlaser fluence at which Si redistribution occurs(Fig. 4). As can be seen in Fig. 4, the Si surfaceconcentration increases with correspondingincreased laser fluence.

Future Work

We have demonstrated that pulsed laserprocessing for the activation of dopants in GaNshows compelling promise. However, additionalprocess development is required before this tech-nology can be applied to actual devices. We willvigorously pursue outside funding for the continua-tion of this work. There are presently opportunitiesfor funding available for GaN microwave poweramplifiers through the Innovative Science andTechnology Program of the Ballistic MissileDefense Organization.

References

1. Mishra, U. K., Y. -F. Wu, B. P. Keller, S. Keller, and S.P. Denbaars (1998), “GaN Based Microwave PowerHEMTs,” Physics of Semiconductor Devices, V. Kumarand S. K. Agarwal, Eds., Narosa Publishing House,New Delhi, India, pp. 878-883.

2. Akasaki, I., H. Amano, M. Kito, and K. Hiramatsu(1991), J. of Luminescence, 48/49, p. 666.

3. J. H. Edgar (1994), Properties of Group III Nitrides,INSPECT, Inst. Electrical Eng., London, UK ISBN 085296 818 3, p. 192.

FY 98 2-35

260 Meyer_qk 7/22/99 5:11 PM Page 2-35

lass Etching


Introduction

Glasses come in many different types andcompositions, with vastly different properties anduses. A relatively new photostructurable glass maybe patterned into microstructures by exposingregions to be etched with deep UV light. Whenproperly annealed, these regions crystallize into aphase of different composition that etches twentytimes faster than the surrounding amorphousmaterial in dilute HF.

The initial surface finish, hidden sub-surfacedamage, and thermal history of a glass may dramati-cally affect the formation of microstructures. Manyprojects at LLNL’s Microtechnology Center (MTC)have been patterning borosilicate glass microstruc-tures with varying degrees of success, but the qual-ity of the patterned microstructures has often beenseverely impaired by the lack of suitable processinfrastructure and understanding of the basic causesof defects.

Our recent work on the patterning of large micro-capillary arrays (50 cm long, 100 µm wide and 10 to100 µm deep) has demonstrated how importantcertain initial conditions are in determining thequality of the etched pattern. Our evidence showsthat non-visible initial scratches and defects play amajor role (sometimes disastrous) in the quality andshape of the etched features.

Three glass treatments were studied as ameans of eliminating or reducing the effects ofsuch defects. The variation of undercut ratios(side cut/depth) from 1.0 (isotropic) to 2 orgreater (anisotropic—fast side etch) is commonlyobserved and poorly understood. This is attrib-uted, in part, to initial glass conditions, but alsoto etching technique (for example, etch composi-tion, orientation, convection, reaction rate, reac-tant diffusion, reaction product removal).

Systematic study of these process parameters isdescribed below.

Progress

Foturan Photostructurable Glass Processing

The recommended wavelength for exposure ofFoturan is 290 to 330 nm, where it absorbsmoderately. Since it does not absorb at 405 nm (ourstandard flood/aligner source), the 235-nm lampwas used. Foturan absorbs strongly at this wave-length, so the depth of the region that is adequatelyirradiated, and therefore crystallized, was stronglydose-dependent, whereas it is weakly so at theprescribed wavelengths.

Samples were irradiated with 2, 4, 8 and16 J/cm2 using a test pattern of lines with spacings

FY 98 2-37

A radically new type of glass, photostructurable glass, has been evaluated for the first time forsuitability for micro-electromechanical (MEMS) structures. The final etch can have an anisotropyratio of up to 10:1 (20:1 if etched from both sides), and etched features may have steep vertical wallangles of 1 to 4°. These initial results indicate that the glass may be very deeply etched, through thewafer if etched from both sides. However, the trench bottom and sidewall roughness of ±2 µm maypreclude some applications with the current level of processing. We have done a multi-parameteretch study on conventional borosilicate glass, since this glass is commonly used in MEMS structuresfor silicon anodic bonding and glass-to-glass fusion bonding. These types of structures are currentlyused in major projects at Lawrence Livermore National Laboratory (LLNL). The results have greatlyimproved the process control to establish undercut ratios, and the control of etch defects, etch-depthuniformity, and etched surface roughness.

Harold Ackler and Stefan P. SwierkowskiElectronics Engineering Technologies Division Electronics Engineering


from 20 to 160 µm, with channel spacings equal tochannel widths, and large rectangular features 200to 1000 µm on a side. After annealing, the pieceswere etched 10 min (Fig. 1). Etch depths measuredwith a stylus profilometer are summarized in Table 1.

Similarly exposed samples were annealed, thencross-sectioned, polished, and etched ~5 min toreveal the depth of crystallized material.Crystallization depth vs dose is summarized for the20- and 40-µm channels in Table 2. The largefeatures were crystallized completely through the0.5-mm wafer for a 16-J/cm2 dose. Only the16-J/cm2 dose appears sufficient to produce crystal-lization deep enough to fully evaluate possible etchaspect ratios, which are approximately 3:1 withsidewall angles of approximately 6°.

The failure to meet figures specified by the manu-facturer are due to the different wavelength usedhere, which alters the process. The surface rough-ness was ±2 µm. A means of smoothing the surfaceis yet to be developed. Finally, the CTE of this glassis approximately 8.5 ppm/°C, making it incompatible

with Si and Pyrex. However, it is CTE-matched to Ti,Pt, Al2O3, and other glasses.

Process Control for Borosilicate Glassesand Initial Defects

The glass defects responsible for the generationof etch defects are believed to be due to either local-ized stress caused by mechanical damage, or chemi-cal inhomogeneity. The former should be presentonly near the surface, and possibly relieved by ther-mal treatments. One group of Pyrex wafers waspolished by a very gentle technique and anotherannealed. A final group was pre-etched in HF toremove defective material. These groups andanother of as-received wafers were patterned andetched with 40-µm channels. The etch defects inchannel walls were counted for known lengths ofchannels and are shown in Table 3.

Annealing was not effective in reducing defectrates. Polishing was very effective in that the defectrate was much lower and the defect size was much


Table 1. Etch depths (µm).

Dose Feature width (µm)(J/cm2) 40 80 120 160 1000

2 20 30 254 20 35 358 45 65 6516 50 65 75

Table 2. Crystallization depths (µm).

Dose Feature width (µm)(J/cm2) 20 40

2 24.5 27.54 47.5 37.58 10016 164

Table 3. Defect rates in treated Pyrex wafers.

Wafer Average Standard deviation cm of 40-µmtreatment (defects/cm) (defects/cm) channel

As-received 47 26 ~ 6Annealed 35 18.6 ~ 6.5Polished 6.8 2.5 ~ 6.5Pre-etched 0.23 0.46 ~ 4

Figure 1. Foturan , etched 10 min at 8 J/cm2.


smaller. However, this method costs approximately$200/wafer through a vendor. The pre-etch wasremarkable at reducing defect rates to almost zero,with very small defects when observed. This processtakes an extra 15 min; however, it leaves pits scat-tered over the entire surface, which are at most250 nm deep. The effects of those pits that intersectchannel edges on device performance have not beenfully considered. They might be sealed up during bond-ing; however, this has not been completely evaluated.

A large set of parametric etch experiments werecarried out on borosilicate (Pyrex) glass using Cr-Auas the etch mask with a varying grating pattern1. Afew selected results are discussed. The quality ofetching is strongly dependent on the etch composi-tion, concentration, mask composition and defini-tion, agitation, and prior history of the glass surface.Data from a frequently used (60 min duration/21 °C)etch is shown in Fig. 2.

These etch experiments show that the absoluteroughness (vs normalized with depth) for Fig. 2 isabout 10 to 40 nm. This is true over a wide range ofconcentrations and depths, and to some extent,composition. Roughness can be dramatically worseif the surface of the glass is damaged, even withnon-visible defects, especially scratches. For veryhigh etch rates, the reaction products can’t beremoved quickly enough, resulting in very poor etchdefinition and roughness.

In Fig. 2, the absolute value of roughnessremained approximately constant at ~30 nm as theHF concentration increased. However, the higher HFconcentrations more rapidly remove the patternedfield photoresist (6 µm thick), which is left over theCr/Au etch mask (1 µm thick) as additional support.

Since this etch is approximately isotropic, it under-cuts the Cr/Au/photoresist mask significantly.

Uniform etch agitation is essential for uniformdepths. The higher agitation needs a stronger maskand the higher the agitation, the more the mask isundercut. For etch times much longer than 60 minin the 22%-HF etch, the mask starts to seriouslydegrade with pinholes or cracking. The 22%-HF etchis a compromise between etch rate, roughness, andmask degradation.

A typical etch result is shown in Fig. 3. The totalmask undercut is about the same as the depth, thatis, 40 µm. The etch rate is reduced in the cornerunderneath the mask near the surface. The reactionrate and product removal is reduced here by limitedconvection, so the glass is locally etched anisotropi-cally and yields a final glass undercut of 5 µm. Thelateral definition of the glass edge appears to signifi-cantly replicate the roughness of the Cr pattern afterit is etched. Care must be take to not over-etch theCr, as it will undercut the Au and result in morejagged Cr edges.

Etch studies taken with small (7 cm) testpieces have yielded significant differences inuniformity compared with very large rectangular(15 cm × 58 cm) plates. The etch-volume tosurface-area-being-etched ratio, combined withthe detailed etch flow patterns during agitation,appears to be very important.

For etching small samples placed vertically alongthe wall of the etch beaker we have etch rates ofabout 0.75 µm/min—independent of zero, medium,or high-spin speeds. For etching small samplesplaced horizontally along the bottom of the etchbeaker, we have etch rates of about 0.89 µm/min—

FY 98 2-39

0

0.5

1.0

1.5

2.0

2.5

HF % vol / acetic acid

0

0.05

0.10

0.15

0.20

0.25

0.30

0.35

0.40

Ro

ugh

nes

s, %

of

dep

th

Etch

rat

e, µ

m/m

in

0 604020

Figure 2. Etch parameters for borosilicate glass.

Figure 3. Etched channels in borosilicate glass, 40 µm deep, ona 250-µm pitch. The 5-µm undercut is key to forming arounded channel.


independent of zero, medium, or high-spin speeds.The etch uniformity across these small samples isvery good—less than 2%.

For large rectangular plates with a longitudinaletch agitation, the etch uniformity of the initialexperiments could be higher, lower, or in the middleby 25%, depending on the agitation conditions. Thelateral etch uniformity was very good—less than2%. By optimizing the agitation conditions for largeplates, overall longitudinal uniformity of etch depthof <4% has been achieved simultaneously withexcellent lateral uniformity.

Using the baseline etch composition of HF 22%vol/acetic acid at 21 °C, the etch rate was reduced to2/3 its value with a temperature of 5 °C, while thesurface roughness may have been reduced slightly.The surface roughness measurements were highlyvariable since they were not averaged much at all.The exception is the roughness of the glass beforeprocessing, and on the unetched surface afterprocessing, and this value was consistently between0.8 and 1.3 nm.

Trying a few experiments with other HF-typeetches including nitric acid, water, and acetic acidwith small amount of surfactant did not yield anydramatic changes in etch rate or surface roughness,except that in some cases the etch mask wasattacked more rapidly. Attempts to use sputtered Moas an etch mask failed because of persistentpinholes and very high stresses in the film that madeit crack during etching. Cr/Au/photoresist hasworked very well as a mask material: e-beam Cr(30 nm) gives excellent adhesion to the glass andthe subsequent Au. Au (1 µm) gives a thick, compli-ant, low-stress layer that is highly etch-resistant.The top layer of photoresist (6 µm) helps to mechan-ically support the Au as the etch proceeds to under-cut the metal mask with significant etch convection


taking place. This type of coating is amenable to lowparticulate defects and very large substrates. Thismasking layer can even be blown dry with nitrogenfor intermediate depth inspection, followed bycontinued etching.

Future Work

Continued study of Foturan® processing willinclude increasing etch aspect ratios and etch unifor-mity, and reducing surface roughness. The effect ofsurface pits generated during pre-etching of Pyrexwafers on device performance and potentially sealingthem during bonding will be studied. The etching ofborosilicate glass seems to yield an initial surfaceroughness of about 20 nm that doesn’t get rapidlyworse with increasing etch depth. The factorscontrolling this roughness are not well understood.

The inhomogenities and impurities in the glasshave not been characterized and should be studiedfurther. For etch depths much more than about75 µm, improved mask materials or etch composi-tions will be needed.

Acknowledgments

We gratefully acknowledge the assistance ofD. Hilken and L. Tarte with the lithography and etch-ing. Support for the microchannel array portion of thegenome sequencing project at LLNL is appreciated.

Reference

1. Swierkowski, S. (1997), “Large Area Lithography,”Engineering Research, Development and Technology,Lawrence Livermore National Laboratory, Livermore,California (UCRL-ID-125472), p. 3-23.

icro-Electromechanical Systems Foundry Interface


Introduction

The Microtechnology Center at LLNL is primarilya research facility for design and development ofmicrostructures and microsystems. We typicallyperform the research needed to develop novelmicrostructures for a variety of programmatic andexternal customers. We generally build prototypes ofthe devices and systems we develop, but are notequipped to provide significant production quanti-ties. We have several customers, however, whowould like small-scale production quantities ofdevices. Recently, several commercial companieshave advertised that they provide foundry service forMEMS, just as foundries have existed for customintegrated circuits for several years.

Most of the customers who would like productionquantities are working on tight budgets and sched-ules, and cannot afford the cost and time delay forus to establish the interface and learn how to workwith foundries. A small investment in establishingthis interface and learning how to work withfoundries will make it much more attractive and effi-cient for the programmatic customers to get theservices they need.

Some foundries have low-cost multi-projectfabrication services if one can use their standardprocess. These have a well-defined interface andare easy to use. They were not addressed by thisproject. We were interested in establishing aprocess for custom fabrication that is more generaland able to satisfy a broader set of our needs. Webegan this project by contacting three possiblefoundries for custom silicon fabrication. They wereidentified from ads in trade journals, personalcontacts, and internet searches.

All three foundries told us essentially that theycan only provide large volume fabrication to be prof-itable. Microelectronics Center of North Carolina(MCNC) gave us a quote for a small run, but it wasapproximately five times higher than our costs tofabricate the same component in-house. We turned,then, to small custom fabrication companies whodid not consider themselves as “foundries” per sebut were more willing to consider the small produc-tion runs that we need. There were three in the SanFrancisco Bay Area whom we identified primarilyfrom personal contacts established over the years.One of these, Nanostructures Inc., has proved to beable to provide the quality and quantities we need at

FY 98 2-41

The goal of this project was to establish a working interface between Lawrence Livermore NationalLaboratory (LLNL) and one or more micro-electromechanical systems (MEMS) foundries so that wecould provide LLNL programs with a one-stop service for research and development of custommicrostructures. Early in the project we found that traditional foundries make most of their moneyfrom volume manufacturing and are, therefore, not interested in the small-scale work that we need.Two small custom fabrication companies were identified, one for silicon MEMS and the other forglass/ceramic component fabrication. The silicon fabrication company proved excellent, and we arebeginning to use their services for programmatic projects. The glass/ceramic fabrication companycould not deliver the quality of components we need, however, and we are going elsewhere for thosecomponents. This project was extremely useful in that we were able to evaluate the capability of twocompanies, narrowed down from a much wider field. Also, we can now provide the componentsneeded by our programmatic customers with confidence.

Michael D. PochaElectronics Engineering Technologies DivisionElectronics Engineering

270 Pocha_qk 7/22/99 5:16 PM Page 2-41


reasonable cost (comparable to fabricating in-house). We have already used their services for oneprogrammatic customer with very favorable resultsand are beginning to perform another run for asecond customer.

In a second technology area, we are increas-ingly interested in fabricating components in non-semiconductor materials like plastic, glass, andceramic. Here we identified a company, AppliedCeramics in Fremont, California, that was willingto do glass and ceramic cutting more quickly andat less cost than others. Applied Ceramics had thepotential not only to perform precision drilling andcutting of substrates, but also to providesubstrates of many different materials.Unfortunately, they were not able to deliver thequality of material needed. But, we found that outwithout having committed any of our customers’funds. This project has been very beneficial toLLNL and will save the programs considerablecost and time in the future. We highly recommendother such projects for technology-base support.

Progress

As mentioned earlier, some foundries offer amulti-project chip where several different designsfrom several customers are all fabricated on a singlesilicon wafer, thus, lowering the cost to eachcustomer. MCNC, for example, offers such a processfor three-layer polysilicon-surface micromachinedcomponents. If you can make your design fit theirstandard process, you can purchase custom MEMScomponents for as low as $2,900 per run. They alsohave other standard processes to choose from. Wehave used this option. However, only about 10% ofour structures can be designed within theconstraints of the standard processes. We need aprocedure for procuring components requiring alarger set of fabrication technologies.

An example we chose as representative of manyof the components we need to fabricate is the siliconoptical microbench. These chips need to have 1) pedestals or wells etched into the silicon surfaceseveral tens of micrometers deep; 2) polysilicondeposited and doped to obtain the correct resistivityfor electrically activated heaters for melting solder;and 3) metal interconnects patterned on both levelsof the steps in the silicon surface created during thepedestal/well etch. This set of steps encompassesthe majority of silicon micromechanical componentswe design and build.

We are also often designing and building compo-nents out of materials other than silicon (glass, plas-tic, ceramic). Here, the primary new process step isthe etching and drilling of holes in these materials.Patterning of plastic is a very broad and complexsubject which was beyond the scope of this project.

Two companies that we know of provide custompatterning of glass and ceramic: Bullen Ultrasonics,Inc., in Eaton, Ohio and Applied Ceramics inFremont, California. We have used BullenUltrasonics in the past, and so chose to try outApplied Ceramics, geographically closer to LLNLand offering lower cost service.

Table 1 below summarizes the results of ourdiscussions and interactions with the companieswe contacted.

In summary, after our extensive search, we foundone company that was able and willing to fabricatethe silicon microbenches and another company toevaluate for etching and drilling of glass andceramic substrates. The silicon microbench fabrica-tion was more successful than the glass and ceramicsubstrate fabrication.

Nanostructures Inc. delivered sil iconmicrobenches, which we were able to deliver toan LLNL program. Our customer was happy toget free parts, but more importantly, pleased thathe can now purchase microbenches, in quantity,at a fixed cost. We are in the process of makingmodifications to the design for the next genera-tion of microbenches to be purchased fromNanostructures Inc.

We have also just placed an order for PCRThermalcyclers, another component that requiresthe same silicon etching and polysilicon resistorformation process. We anticipate out-sourcing thesewell-defined and relatively simple designs, ulti-mately saving LLNL significantly more than the costof this project.

Our experience with Applied Ceramics was not assuccessful. We contracted with them to deliver tenpolished Al2O3 ceramic wafers with holes drilled ina prescribed pattern, and to drill a similar pattern ofholes in Pyrex wafers that we supplied to them. Theceramic wafers were delivered with surfaces thatwere too rough for our applications and the holes,although placed correctly, had very rough andchipped edges. Also, the glass wafers had severelychipped holes which are unacceptable for our appli-cations. We will, therefore, continue to get glass andceramic substrates from the more expensivesupplier, Bullen Ultrasonics, Inc.


270 Pocha_qk 7/22/99 5:16 PM Page 2-42


FY 98 2-43

Table 1. Summary of foundry research.

Silicon MEMS foundries

Microelectronics Center of N. Carolina (MUMPS) Multi-project run. Bid - $2900 for designs that use 3021 Cornwallis Road standard three-level polysilicon-surface micromachine Research Triangle Pk., NC 27709 processing.(919) 248-1800

Full custom processing such as the silicon microbench; only interested if production lots are sufficiently large to be prof-itable to produce. Bid - $53,000.

Standard Microsystems Corp. Set up for volume production of silicon components; only 35 Marcus Boulevard interested at the 100 to 1000 parts per week level. No bid.Hauppauge, NY 11788(516) 435-6961

MEMStek Products, LLC Specialize in microfluidic delivery systems such as pumps and2111 SE Columbia Way valves. Not interested in general MEMS. No bid.Suite 120Vancouver, WA 98661

Small custom fabrication

MicroFlow Inc. Initially interested in bidding on our project, but were just 6701 Sierra Court acquired by a large custom house, Input/Output Inc., Houston,Dublin, CA 94568 Texas. While we were in discussions, the new management (925) 828-9650 made the decision that this project was too small to warrant

further effort, so discussions ended. No bid.

TiNi Alloy Company Specialize in shape memory alloy actuated devices using Ti and1621 Neptune Drive Ni alloys. Had some initial interest in silicon fabrication to San Leandro, CA 94577 expand their capability, but while we were discussing our (510) 483-9676 requirements, decided to concentrate on their core business.

No bid.

Nanostuctures Inc. Interested in our job. Bid $9500.3070 Lawrence Expressway Successfully fabricated microbenches.Santa Clara, CA 95051(408) 733-4345

Non-silicon fabrication

Applied Ceramics Delivered glass and ceramic substrates with holes in specified850 Corporate Way locations. Bid - $4150.Fremont, CA 94539 Fabricated parts, but unacceptable quality.(510) 249-9700

270 Pocha_qk 7/22/99 5:16 PM Page 2-43



Even this negative experience was usefulbecause it confirmed that we were getting ourglass and ceramic substrates from the most costeffective source.

Future Work

This was an excellent technology-base project.We recommend that each year a small amount oftechnology-base funding be set aside for the identifica-tion and evaluation of vendors to supply well-defined,

non-research components. We can concentrate ourmore expensive resources (people and equipment)on the research we do best, but still supply ourcustomers who need them a significant quantity ofproduction components.

Acknowledgment

The author would like to thank H. Ackler at LLNLfor conducting the non-silicon fabrication part ofthis project.

270 Pocha_qk 7/22/99 5:16 PM Page 2-44

Hydrogel-Actuated Implantable Sensor


IntroductionHydrogels are a class of polymers that have a

high capacity for water absorption. Typical massincreases for the hydrated state of such polymers,for example, poly-hemas (PHEMAs), may be 10 to100 times, and swelling volume ratios of up to 30%can be achieved. We are interested in suitably modi-fied polymers that, in the hydrated state, exhibitreversible swelling in response to a change in pH,osmolarity, temperature, pressure, or the concentra-tion of some analyte of interest. Enzymes can beembedded in the hydrogel to create specificity to agiven analyte.

Currently, hydrogels are commercially used inabsorbents, thickeners, dilators, and as osmoticpumps. However, few researchers have usedhydrogels as an actuator for micromechanicalstructures. With the swelling volume achievableby these polymers, significant mechanical workmay be accomplished.

Progress

We have demonstrated a micro-electromechanicalsystem (MEMS) sensor that is sensitive to theconcentration of an ionic solution, and a prototypepassive resonant circuit that can be used for remoteinterrogation of the device. The sensor is actuatedby a biocompatible PHEMA hydrogel that swells orshrinks in response to variations in concentration ofan ionic solution. It consists of a silicon cell thatencapsulates the hydrogel between a deformablemembrane and a liquid-permeable mesh chip(Fig. 1).

As the hydrogel changes dimensionally inresponse to an analyte in solution, it flexes adeformable, conducting membrane that forms oneplate of a parallel plate capacitor. The capacitivevariation can be monitored by incorporating thehydrogel capacitor into a resonant LC circuit, suchthat a change in analyte concentration is reflectedby a shift in resonant frequency of the circuit.

FY 98 2-45

Amy W. Wang, Abraham P. Lee, and Charles F. McConaghyElectronics Engineering Technologies DivisionElectronics Engineering

Christopher B. Darrow, Stephen M. Lane, and Aleksandr Gilman Medical Technologies Program

Joe H. Satcher, Jr.Chemical Sciences DivisionChemistry and Materials Science

With this project we seek to demonstrate hydrogel-based transduction and telemetry technologiesthat could lead to a new class of implanted medical sensor devices. An implanted sensor is useful forany patient who must monitor a particular blood analyte and, thus, requires frequent extraction ofblood samples. Examples include electrolyte and pH measurements (potassium, sodium, calcium) formany chronically ill patients, urea and potassium for dialysis patients, immuno-suppressant drugs fortransplant and AIDS patients, anti-coagulants for many heart and stroke patients, and glucose fordiabetic patients. We envision a wristwatch-sized, externally-powered component worn by thepatient, which interrogates a passive, implantable element.

275 Wang_qk 7/22/99 5:19 PM Page 2-45


This design can be used in an implantablesystem in which a signal is sent transdermally,sweeping a range of frequencies to interrogate theimplanted device and detect variations in resonantfrequency. The signal can be transmitted orreflected back to the interrogating circuit, which ismounted at the skin surface in a readout packagesimilar to a wristwatch.

Silicon Cell

The silicon cell encapsulating the hydrogelconsists of two silicon chips bonded together. Onewall of the cell is a deformable membrane formedfrom a 2-µm-thick biocompatible NiTi-based filmsputtered on a 0.5-µm silicon nitride membrane.Nitride membranes are formed by silicon etching inpotassium hydroxide (KOH). We found that NiTi-based membranes are preferable to gold-coated sili-con nitride membranes in durability, since contrac-tion of the polymer during drying results in breakageof the gold-nitride membranes. The NiTi-basedmembranes were used successfully and found to bequite robust.

The opposite wall of the cell consists of a rigidmesh made up of a 50-µm-thick silicon chip with100- to 200-µm-size holes etched through its thick-ness. The mesh allows exposure of the hydrogel tothe ambient analyte solution, while constraining thehydrogel. Fabrication of the mesh involves double-sided exposure of the mesh area and mesh holes ona silicon wafer, then simultaneous KOH silicon etch-ing from both sides of the wafer, thinning it down toa thickness of 50 µm and etching through the holes.

Hydrogel Testing

Both uncross-linked and cross-linked PHEMAhydrogels were tested for swelling in response to the

concentration of a calcium nitrate solution. Uncross-linked PHEMA gels were dissolved in methanol,pipetted into the silicon cell in liquid form, thendried before rehydration in water.

Cross-linked gels were dried in a nitrogen envi-ronment before rehydration. Gel swelling wasobserved for both types of gels, but cross-linkedpolymers exhibited more lateral swelling, ratherthan out-of-plane swelling that would displace theflexible membrane.

Uncross-linked PHEMA displayed good out-of-plane displacement, and increased swelling wasobserved with increasing salt concentration.

Optical membrane displacement measurementsshowed 4 µm, 20 µm, and 30 µm peak displace-ments for calcium nitrate concentrations of 0.5 M,1 M, and 3 M, respectively. Figure 2 shows anexample of an optical measurement of membranedisplacement (~30 µm) due to polymer swelling. Athigh salt concentration, the polymer began toextrude through the holes of the mesh chip due to“untangling” of the polymer strands. A thin layer of“stiffer” polymer used between the PHEMA and themesh chip reduced the amount of extrusion. We haveyet to determine hysteresis effects of the hydrogel.

Interrogation Circuitry

We have also built a resonant antenna circuit inwhich an inductor coil remotely probes a passive LCcircuit. A network analyzer was used to sweep theinterrogation frequency and measure the powerreflected back through the inductor to indicate theresonant frequency (minimum reflected power) asshown in Fig. 3. We were able to interrogate thepassive circuit with the inductor 5 mm away from thecircuit, a distance that is sufficient for a sensorimplanted directly beneath the skin. From opticaldisplacement measurements, we predicted capacitive


4.0 mm

4.0 mm

Figure 1. Diagram of hydrogel-actuated silicon MEMS cell.Hydrogel swelling in response to an analyte causes deflectionof a conductive membrane and a resultant change in capaci-tance of a parallel plate capacitor.

Figure 2. Optical measurement of membrane deflection due tohydrogel swelling.

Hydrogel swelling force Silicon

NiTiSiliconnitride

Analyte solution

Hydrogel

Silicon

275 Wang_qk 7/22/99 5:19 PM Page 2-46


changes on the order of 1 pF due to polymerswelling. Using our antenna circuit, we were able todetect 1-MHz shifts in an 18-MHz resonant peak forincremental changes of 1 pF (Fig. 3).

In summary, we have demonstrated that signifi-cant shifts in resonant frequency can be observed asa result of hydrogel swelling, and that it is feasibleto implant a passive circuit and use telemetry tointerrogate the implanted element.

Acknowledgments

The authors gratefully acknowledge H. Ackler andP. Krulevitch for their ideas, discussion, andprocessing advice, and P. Ramsey for the depositionof the NiTi films.

FY 98 2-47

Ref

lect

ed p

ow

er (

dB

M)

7pf

6pf

-0.35

-0.30

-0.25

-0.20

-0.15

-0.10

-0.05

0.00

Frequency (MHz) 1 0 1 5 2 0 2 5

8pf

Figure 3. Reflected power from the resonant circuit. Resonanceis indicated by minimum reflected power. Capacitive variationsof 1 pF result in approximately a 1.5-MHz shift in an 18-MHzresonance frequency.

275 Wang_qk 7/22/99 5:19 PM Page 2-47

ambda Connect: Multi-Wavelength Technologies for Ultrascale Computing

Microtechnology

Introduction

Ultrascale computing—the integration of largenumbers of processors into a single, highly capablemulti-processor system—is currently of great inter-est in several government programs. These systemscontain 100s to upwards of 1000 CPUs, to attaincomputing capabilities from 100 GFLOPS to 100TFLOPS and beyond.

The provision of fast data communications withinsuch systems poses a significant challenge. Effectivecommunication is currently hampered by the band-width, latency, and congestion characteristics of theelectronic fabric used to interconnect the manysystem-processing and memory elements. Thiscommunications bottleneck can substantiallydegrade computational performance, significantlycomplicate programmability, and cause inefficientuse of costly memory resources.

The recent emergence of byte-wide optical inter-connects, which use linear arrays of multi-modeoptical fibers in a ribbon cable assembly, hassubstantially improved the cost and performance ofGbyte/s point-to-point communications. To fullyleverage this technology, however, the latency andcongestion issues with distributed electronic switch-ing must be overcome.

We have recently proposed source-routed switchingin the optical domain using a wavelength-switchingmechanism. We have shown that this approach canyield highly capable switches that support 100s of

Gbyte/s ports and yield minimal internal congestionand latency. Instruction-level simulation has demon-strated the advantages of this approach.1

To implement this multi-wavelength optical inter-connect requires the addition of multi-wavelengthcapability to the existing byte-wide optical intercon-nect. The required components, which include opti-cal transmitters capable of fast wavelength tuning,and fixed optical filters differ substantially fromexisting telecommunications fiber optics. All compo-nents must be compatible with existing multi-modefiber ribbon cables, and provide compact formfactors suitable for populating electronics boards.

This project aims at developing prototypes ofthese components, to investigate the feasibility ofimplementing byte-wide, multi-wavelength opticalinterconnects for ultrascale computers.

Progress

Multi-Wavelength Transmitters

We have developed prototype modules with thecapability of transmitting on any of two or fourselectable wavelengths. The module wavelength isselected by directing current into different lasers,each laser emitting at a different wavelength, so thatcurrent switching is used to provide fast wavelengthswitching on a timescale of the laser modulationbandwidth (1 to 2 ns switch time). Each modulecontains one linear array of 10 laser diodes for each

FY 98 2-49

Byte-wide, multi-wavelength optical interconnects can substantially improve the performance ofthe system interconnects used in ultrascale computing platforms (“supercomputers”), leading tosubstantial improvements in overall system performance for applications that require global commu-nications within the supercomputer. This year, we have developed the first generation of optical hard-ware required for such interconnects: byte-wide multi-wavelength transmitters and wavelength filters.The performance of these prototypes demonstrates the viability of our approach.

Robert J. Deri, Michael C. Larson, and Michael D. PochaElectronics Engineering Technologies DivisionElectronics Engineering

Mark E. LowryDefense and Nuclear Technologies

280 Deri_qk 7/22/99 5:29 PM Page 2-49

Microtechnology

wavelength. Each array element can be indepen-dently addressed (modulated) by a current drive, sothat the module can transmit 10 independent datastreams comprising the 8 bits in an electronic dataword of 1 byte capacity, plus clock and framing.

Our modules use arrays of vertical cavity,surface-emitting lasers (VCSELs), which we havefabricated from AlGaAs semiconductor layers grownby molecular beam epitaxy. Our VCSEL design usesan oxidized AlAs layer to control the “active region”of the laser which is pumped by the applied currentdrive, to achieve lasing threshholds of 2 mA fordevices of 5 to15 µm oxide aperture.

The VCSELs are sized to emit in multiple trans-verse modes (1.5 nm spectral extent), to minimizespeckle noise during multimode fiber transmission.These devices provide milliwatts of optical powerwhen pumped with <10 mA at ≤3 V, conditionscompatible with conventional CMOS electronic driveelectronics. Figure 1a shows a typical VCSEL opticaloutput characteristic as a function of drive current.

We use a separate VCSEL array chip for eachselectable wavelength in our modules, with eachchip originating from a separately processed AlGaAswafer. This approach enables us to select an arbi-trary spacing between system wavelengths, andenables the system wavelengths to span a very widespectral range (many 10s of nanometers). This isattractive because it allows a moderate channelseparation (8 to 10 nm) that minimizes issues asso-ciated with wavelength registration, uniformity, anddrift between different photonic components withinthe interconnect.

This year, we demonstrated a dual-wavelengthtransmitter module that can launch light of either815 or 830 nm into all 12 fibers in a ribbon cable.The optical signals are independently addressable,

and originate from two VCSEL chips with approxi-mately 6 µm oxide apertures.

Figure 1b shows our ceramic pin-grid-array(cPGA) module, which contains two laser chips.Each laser chip provides a different wavelength, andthe emitters from both chips are coupled into asingle fiber using passive optical alignment to astandard ribbon cable.

The dual-wavelength module exhibits good perfor-mance, as shown by the optical spectrum and400 Mbit/s eye pattern of Fig. 2. The spectrumshows simultaneous emission into a single fiberfrom two wavelengths, with signal-to-noise ratios of30 dB. All laser elements are functional, and deliver>2 mW of optical power into the fiber cable.

We have verified that the packaged lasers aresuitable for high speed operation by observing theirperformance under high-speed modulation.

Our cPGA package exhibits resonance-freeperformance to 300 MHz when mounted with appro-priate external impedance matching resistors. Weanticipate that the module frequency response canbe extended to the VCSEL device limit of severalGHz by using suitable internal terminations anddriver electronics, and possibly by transitioning to ahigher-speed ceramic package.

We evaluated our module’s transmission perfor-mance in a link using a commercial 400 Mbit/sreceiver which performs some reshaping. The result-ing eye pattern (Fig. 2) exhibits an open eye andmeasured bit error rates below 10–14 at 400 Mbit/sper fiber for a pseudo-random bit stream ofsequence 223-1 bits.

We extended our optical packaging approach to afour-wavelength transmitter module by using aproprietary, optical superstrate assembly. Thisassembly combines optical signals from four VCSEL


Out

put

po

wer

(m

W)

Current (mA)5

7

6

5

4

3

2

1

015 200 10

Alignment pins Cable connectorLaser chips

(a) (b)

0utput

Figure 1. Dual-wavelength trans-mitter module: (a) light output vsdrive current forVCSEL device(unpackaged); (b) overall ceramicPGA module, show-ing alignment to afiber ribbon cable.The cable connec-tor is 6.4 mm wide.

280 Deri_qk 7/22/99 5:29 PM Page 2-50

Microtechnology

arrays, operating at wavelengths 823, 843, 971, and986 nm. Figure 3 shows the complete four-wave-length module in its cPGA package, and its outputspectrum when all wavelengths are simultaneouslyactivated. The module uses a similar guide pinapproach for connecting to ribbon cable assemblies.

Byte-Wide Filter Modules

We have developed optical filter modules based onthe packaging of thin-film interference filters within ahousing comprised of ribbon cable connector

ferrules.2 The advantages of this approach are that1) the module is assembled primarily by passivealignment using interlocking guide pins, and 2) themodule automatically mates to ribbon cable connectors.

This year, we extended our filter approach tonarrow band filters and to improved transmissioncharacteristics for enhanced system channel spac-ing and wavelength tolerances.

Table 1 summarizes the performance of modulesusing single-cavity Fabry-Perot filters. The datashow that acceptable insertion loss can be achievedfor filter bandwidths as small as 4 nm. Modules withthe broader (14 nm) passband have been cascadedin series to improve cross-talk suppression and toachieve somewhat narrower passbands. Themodules exhibit a fiber-to-fiber wavelength unifor-mity (±1 nm) sufficient for channel spacings ofseveral nanometers.

Filter modules with a flatter top and sharperskirts than are obtainable from the Lorentzian line-shape associated with single cavity designs are valu-able, because they improve system tolerances tovariations and drifts in the wavelengths of individualtransmitter and filter modules, and enable tighterspacing of wavelength channels.

FY 98 2-51

Table 1. Single cavity Fabry-Perot filter characteristics.

Filter bandwith 14 nm 4 nm

Number of layers in Bragg reflector 12 20Packaged insertion loss (dB) average, <best> 1.5 <0.8> 2.7 <1.9>Unpackaged insertion loss (dB) (collimated beam) 0.76 1.1Cross-talk suppression (dB) -11 (no cascade) -19

-60

-55

-50

-45

-40

-35

-30

-25

Wavelength (nm)

Rel

ativ

e o

utp

ut p

ow

er (

dB

)

1000950900850800

Figure 3. Output spectrum of four-wavelength module incPGA. Note that the spectrum is displayed on a logarithmic(decibel) scale.

-70

-60

-50

-40

-30

-20

-10(a)

(b)

790 800 810 820 830 840 850 860 870

Rel

ativ

e in

ten

sity

(d

B)

Wavelength (nm)

Rec

eive

r p

ho

tosi

gn

al (

V)

0 543210.10

0.15

0.20

0.25

0.30

0.35

0.40

Time (ns)

Figure 2. Dual-wavelength module spectrum (a), and high-speed signalling performance (eye diagram)(b). Note that thespectrum is displayed on a logarithmic (decibel) scale.

280 Deri_qk 7/22/99 5:29 PM Page 2-51

Microtechnology

We have demonstrated such a filter module,based on a multi-cavity interference filter design.Its reponse (Fig. 4) shows a flatter top andsubstantially sharper filter skirts than are achiev-able with single cavity devices, and exhibits cross-talk suppression better than 30 dB. Filter skirtsharpness can be quantified by the ratio of thefilter’s 3 dB bandwidth to 20 dB bandwidth. Thisratio is 0.48 for the device of Fig. 4, three timessharper than our single cavity devices. This deviceis suitable for channels spaced 8 to 10 nm apart,and provides a 5-nm tolerance window for compo-nent wavelength variations (at 1 dB excess loss).

Future Work

These results demonstrate the feasibility of high-performance optical components for byte-wide,multi-wavelength optical interconnects, showingthat compact, high-performance transmitter andfilter modules can be realized for systems using fourwavelength channels.

Our goal for next year is to demonstrate compo-nents enabling systems with higher wavelengthcount, since earlier simulations showed that eightsystem wavelengths are required for several inter-esting multiprocessing applications.2 We also intendto develop additional passive routing devices forbyte-wide interconnects, including N-to-N starbroadcast components.

Acknowledgments

This project has benefited substantially fromthe efforts of M. Emanuel, H. Garrett, W. Goward,R. Patel, and H. Petersen of LLNL, and fromcontributions by Prof. C. Gu (University ofCalifornia, Santa Cruz), and Prof. R. F. Drayton(University of Minnesota).

References

1. De Groot, A. J., and R. J. Deri (1996), Proc. ThirdIntl. Conf. Massively Parallel Processing usingOptical Interconnections, IEEE, October.

2. Deri, R. J., and S. Gemelos (1998), “Simple fabrica-tion of WDM filters for byte-wide, multimode cableinterconnects,” Integrated Photonics Research Conf.,Victoria, Canada, April.


-50

-40

-30

-20

-10

0

860840820800780760

Tran

smis

sio

n (

dB

)

Wavelength (nm)880

Figure 4. Transmission response of multi-cavity filter module.

280 Deri_qk 7/22/99 5:29 PM Page 2-52

Kenneth L. Blaedel, Center Leader

Center for Precision Engineering

The mission of the Center for PrecisionEngineering at Lawrence Livermore NationalLaboratory (LLNL) is to ensure that programs haveavailable an adequate base of high-precision designand manufacturing technology, not necessarily resi-dent at LLNL, to help solve their critical future prob-lems.

Our specific goals are 1) to develop an understand-ing of fundamental fabrication processes and themodels that reflect that understanding; 2) to advancemethods of the design of machinery that incorporatethose fabrication processes; and 3) to maintaincontinuing relationships among our colleagues inindustry, government and academia that promote ourcollective capabilities in precision engineering.

In support of these goals, three projects arereported here that bring either higher precision orlower cost-of-precision to the manufacturing chal-lenges that we face over the next few years.

The first project, “Micro-Drilling of BerylliumCapsules,” has seen significant advance. Last year’sconclusion was that the evolutionary changes incommercial capability could not be expected to laser-drill holes small enough and precise enough for futureNational Ignition Facility (NIF) capsules. This year’sresult, which included holes made with a femto-second laser, holds promise for being able to do so.

The second report, “A Spatial-Frequency-DomainApproach to Designing Precision Machine Tools,”presents a new view of how we can design machinetools and instruments to make or measure partsthat are specified in terms of the spatial frequencycontent of the residual errors of the part surface.

This represents an improvement in our ability and areduction in cost to design manufacturing machinesin comparison to using an “error budget,” a designtool that saw significant development in the early1980s, and has been in active use since then.

The third project, “Precision Grinding ofMicrofeatures in Brittle Materials,” demonstratesour ability to develop high-precision manufacturingprocesses and then convey them to commercialindustry, which can then supply that technology forhigh production.

In addition to conducting the three projectsabove, the Center for Precision Engineering holdsmembership in two academic consortia, allowing usinsight into broader areas of precision engineeringthat we cannot pursue ourselves.

Looking to the future of precision engineering atLLNL, we have drawn two conclusions. First,conducting the business of LLNL will require machin-ery capable of material removal, deposition, andmetrology to produce components and assemblies toatomic-level dimensional tolerances. Second, signifi-cantly reducing the cost of precision for componentmanufacture and for assembly of precision productswill actually enable many LLNL projects. It is inprojects such as NIF that the expenses of precisionmanufacturing can defeat big physics.

With the focus of this year’s projects on creatinghigh-precision processes and instruments at accept-able cost, we think the Center for PrecisionEngineering has materially contributed to LLNL’sability to field small- and large-scale science.


Center for PrecisionEngineering

3


Contents

3. Center for Precision Engineering

OverviewKenneth L. Blaedel, Center Leader

Micro-Drilling of ICF CapsulesSteven A. Jensen and Brent C. Stuart ........................................................................................................3-1

A Spatial-Frequency-Domain Approach to Designing Precision Machine ToolsDebra A. Krulewich ...................................................................................................................................3-3

Precision Grinding of Brittle MaterialsMark A. Piscotty, Kenneth L. Blaedel, Pete J. Davis, and Pete C. Dupuy....................................................3-9


icro-Drilling of ICF Capsules


Introduction

This report serves as an update to “Micro-Drillingof ICF Capsules.”1 In that report the commercialcapabilities of micro-drilling small holes were inves-tigated for limits of precision, quality, and attainableaspect ratios. The motivation behind the investigationwas to determine the feasibility of drilling small holesin the National Ignition Facility (NIF) fusion capsules,suitable for filling with the intended fuel mixture.

The report concluded that although the micro-drilling of holes in the fusion capsules is feasible,the commercial sector currently cannot producethese holes. SEM photos of commercially drilledholes showed an excessive amount of thermaldamage (that is, dross, re-melt, and thermal crack-ing) due to laser pulse lengths of relatively longdurations (nanosecond and longer). Melt-expulsion,and not evaporative ablation, largely dominates themechanism behind commercially drilled holes.

Furthermore, most commercial companies are notequipped with the appropriate lenses and laser set-upto drill holes smaller than 5 µm because currentdemand is limited. It was speculated in the FY-97report,1 based solely on published data, that laserswith pulse lengths in the femtosecond to 0.1 ps rangecould eliminate much of the thermal damage andpotentially produce small enough holes to meet strin-gent NIF requirements.

This update outlines the preliminary efforts usingshort-pulse (≈100 fs) lasers to potentially drill holesin the ICF capsules.

Progress

The ICF capsules will be made of doped beryl-lium (Be) having an ablator shell thickness between100 and 150 µm. Therefore, the goal of the prelimi-nary studies was to drill 5-µm or smaller holesthrough 125-µm-thick Be foil. In our experimentsthe Be foil was mounted on a xyz-translation stagein a vacuum chamber pumped down to 25 mTorr.The experiments were performed using a 1-kHz,120-fs, Ti:Sapphire short-pulse laser system.Numerous combinations of spatial filtering, focuslens, f-number, and wavelength were tried and thebest focal spot obtained was 5-µm 1/e2 diameter(where spot size is defined as the distance at whichthe Gaussian beam intensity has dropped to 1/e2 =0.135 times its peak value).

To achieve this, the laser output was spatiallyfiltered, frequency doubled to 413-nm, spatiallyfiltered again, and then focused with a 25-mm focallength GRIN lens (LightPath) at approximatelyf-number = 5. Both the quality of the BBO doublingcrystal (λ/2 surface) and the choice of focusing lensleave room for reducing the focal spot size.

SEM images, after ultra-sonic cleaning, of one ofthe smallest holes obtained are shown in Fig. 1. Theentrance diameter is approximately 6.5 µm, whilethe exit diameter is approximately 3 µm. This gives ataper angle of approximately 1.6° which is signifi-cantly better than the 4 to 12° that was seen incommercially drilled holes. This hole was drilledwith approximately 1.8-µJ pulses (18J/cm2) at 1 kHzin 12 s.

FY 98 3-1

Further studies are reported on micro-drilling of holes in fusion capsules. This update outlines thepreliminary efforts using short-pulse (≈100 fs) lasers to potentially drill holes in the ICF capsules.

Steven A. JensenManufacturing and Materials Engineering DivisionMechanical Engineering

Brent C. StuartLaser Science and TechnologyLaser Programs

305 Jensen_qk 7/22/99 5:33 PM Page 3-1


These preliminary studies are encouraging. Thediameter of the heat-affected zone is much smallerthan commercially drilled holes. There were no visi-ble signs of thermal cracking and the re-melt aroundthe walls was minimal. Furthermore, the taper angleassociated with the hole is much smaller than hasbeen seen in commercially drilled holes. It is specu-lated that the focal spot size can be further reducedto approximately 2 µm in diameter by using asmaller f-number and possibly third harmonics. Withsome fine-tuning of parameter settings these holescould potentially meet NIF standards.

Having performed these preliminary studies andprior to further studies, the question of “how smallis small enough” remains to be answered. Initialindications suggested that entrance hole diameterson the order of 1 to 2 µm in diameter would be smallenough. However, studies need to be performedwhich ultimately back up these initial estimates andquantify the largest allowable capsule perturbationthat would affect the hydrodynamic stability andhinder a symmetrical implosion. Whether this workis done experimentally or through simulation, it isnecessary to validate the continued efforts of reduc-ing and refining the micro-drilled hole.

Assuming that a 1- to 2-µm hole would be suffi-ciently small, there are still issues that must beresolved if this approach to capsule filling is tosucceed. The hole will need to be sealed shut oncethe capsule is filled with the intended fuel mixture ofdeuterium-tritium (DT). The sealing process, like thedrilling process, will have to minimally affect theintegrity and surface quality of the capsule. Workneeds to be done to determine the extent and size ofthe area around the hole affected by the sealingprocess. Ideally it would be desirable to use thesame laser set-up to seal the holes as was used todrill them. This might be accomplished by reducingthe intensity of the beam and/or using longer pulsedurations to sinter the hole shut. The feasibility ofthis approach also remains to be studied.

In addition to the sealing process, a micro-polishing process may be necessary to smooth overablator shell perturbations due to the sealingprocess. This would also help to ensure a surfaceroughness that meets specifications. Currently, thespecification for surface roughness is on the orderof 10 nm rms or less, and efforts thus far have notproduced capsules with a surface roughness lessthan 50 nm rms. Micro-polishing of the capsulesneeds to be studied, since this process may beneeded even if diffusion filling is adopted as theapproach to capsule filling.

Reference

1. Jensen, S. A., “Micro-Drilling of ICF Capsules,”Engineering Research, Development and Technology,Lawrence Livermore National Laboratory, Livermore,California (UCRL-ID-129204).


Figure 1. SEM images of smallest holes obtained: a) entrancehole; and b) exit hole.

(a)

(b)

(x3000) 10 µm

(x3000) 10 µm

305 Jensen_qk 7/22/99 5:33 PM Page 3-2

Spatial-Frequency-Domain Approach to Designing Precision Machine Tools


Introduction

Increased precision in manufacturing is beingdemanded by Lawrence Livermore NationalLaboratory (LLNL) Programs in areas ranging fromNIF optics manufacturing and ICF target positioning,to the production and alignment of optics for EUV lith-ography. Other LLNL areas that drive unique require-ments for precision include the machining of diffrac-tive optical systems, the fabrication of ICF targets,and the assembly/packaging of fiber optical systems.

The precision-to-cost ratio is another metric thatrelates to a wide variety of industrial mechanicalsystems, such as automotive engine components,but has a special significance at LLNL where anincreased interest in tighter tolerances is matchedby the need to lower program costs. Minimizingtechnical risk while maintaining precision is acomplementary issue that defines manufacturinggoals for programs that cannot tolerate yield factorsless than 100%, such as in fabricating componentsfor the nuclear weapons program.

This project presents an opportunity to signifi-cantly improve the foundation that underlies ourprecision engineering expertise: the process offormulating an error budget for a manufacturing,positioning, or measurement system. Error budgets

provide the formalism whereby we account for allsources of uncertainty in a process, and sum themto arrive at a net prediction of how “precisely” amanufactured component can meet a target specifi-cation. The error budgeting process drives decisionsregarding the conceptual design of the system andchoice of components and subsystems, and enablesa rationale for balancing precision (performance),cost, and risk.

The principles of designing precision instrumentsfor meeting challenging tolerance requirements havea rich history.1 Likewise, the methodologies foranalyzing the errors in experimental data andperforming differential sensitivity analyses are well-documented.2,3 Yet the first clear formalization oferror budgeting applied to precision engineeringappears to originate in the analysis by R. Donaldsonduring the design of the Large Optics DiamondTurning Machine at Lawrence Livermore NationalLaboratory (LLNL).4 Donaldson’s formalism is refer-enced in current textbooks5 and is the basis forsubsequent machine designs at LLNL.6

Figure 1 shows flowcharts for both the conven-tional and the new error budget procedure and howthey differ. The upper portion of Fig. 1 showsDonaldson’s flowchart illustrating the mapping oferror sources onto part geometry.

FY 98 3-3

The aim of this project is to develop a methodology to design machines used to manufacture partswith spatial-frequency-based specifications, thus reducing risk while maintaining accuracy. Using inerror budget, we are able to minimize risk during the design stage by ensuring that the machine willproduce components that meet specifications before the machine is actually built. Minimizing the riskwhile maintaining accuracy is a key manufacturing goal for programs that cannot tolerate yieldfactors less than 100%, such as the nuclear weapons program. Current error budgeting procedureprovides no formal mechanism for designing machines that can produce parts with spatial-frequency-based specifications. However, recent specifications for advanced optical and weapons systems arebeing posed in terms of the continuous spatial frequency spectrum of the surface errors on themachined part. Based on these requirements, it is no longer acceptable to specify tolerances in termsof a single number that spans all temporal and spatial frequencies. During this project, we willdevelop a new error budgeting methodology to aid in the design of new machines used to manufac-ture parts with spatial-frequency-based specifications.

Debra A. KrulewichManufacturing and Materials Engineering DivisionMechanical Engineering

310 Krulewich_qk 7/22/99 5:35 PM Page 3-3


The first step of the conventional error budgetis to identify the physical influences that generatethe dimensional errors that propagate through themachine tool. These include effects such as ther-mal gradients and temperature variability, bear-ing noise, fluid turbulence in cooling passages,and way non-straightness.

The second step is to determine how this sourcecouples to the machine. A coupling mechanismconverts these physical influences into a displace-ment that has a direct influence on machine perfor-mance. An example of a coupling mechanism is thethermal expansion that may transform a time-varyingheat source in the vicinity of the machine into amachine way distortion. These displacements repre-sent dimensional changes in the system. A singlepeak-to-valley number is usually used to quantifythe dimensional changes, not differentiatingbetween the spatial frequency content of the error.

The next step is to sum all the contributingerrors using an appropriate combinatorial algorithm. Literature suggests a variety of combi-natorial algorithms.7

The last step in the error budgeting procedure isto transform these errors into the workpiece coordi-nate system. To convert these machine displace-ments into the errors that would reside on the work-piece surface in the directions of interest, we mustconsider the tool path (feed rates and spindlespeeds, for example).

The output from this procedure is a singlenumber predicting the net error that would result ona machined workpiece. We would then compare thisnumber to the part specifications. If the predictionmeets target specifications, we would accept themachine design under evaluation. If the predictiondoes not meet specifications, we would evaluatemethods to improve this design by observing which

sources are the dominating contributing errors. Inthis way we can evaluate the cost vs accuracy ofdifferent candidate designs.

If improvements could be made to an existingdesign, we would make those changes to the errorbudget and reevaluate the net error. If the modifi-cations were not practical, we would thenconsider an entirely new design, or possiblyreevaluate the specifications.

Progress

The lower portion of Fig. 1 shows the new errorbudget approach. The first two steps, identifying thesources and how they couple to the machine, areidentical and are explained in the previous section.However, the new approach differs in the next step,where the elemental errors are converted into thefrequency domain. The next step is to combine theerrors in the frequency domain. The combinatorialrule is a completely new algorithm with a statisticalfoundation. These steps are explained below.

Figure 2 displays a block diagram of the machin-ing process. During cutting, an instantaneousamount of material is removed, which createsforces. The ratio between the cutting force andamount of material removed is the material removaltransfer function. These cutting forces combine withforces induced by the machine errors. The machinestructure responds with displacements that ulti-mately result in errors on the machined part.

The conventional error budgeting approach doesnot consider the dynamics of the material removaltransfer function. In other words, the conventionalapproach assumes that the forces are directlyproportional to the amount of material removed, sothe cutting process doesn’t damp or amplify theerror sources at certain frequencies. Our proposed


Combinatorial algorithm

New approach

Conventional approach

Mapping to workpiece coordinate

systemSpatial frequency

content of residual errors in part coordinate system

Pow

er o

r am

plit

ude

Spatial frequency

Number (net RMS error or peak-to-valley error)

Transform to spatial frequency

domain

Combinatorial algorithm

Material removal transfer function

Mapping to workpiece coordinate

system

Identify physical

error sources

Couple to

machine

Figure 1. Flowcharts for both the conventional and the new error budget.



approach considers the dynamics of the materialremoval transfer function. The last step is to trans-form the errors into the part coordinate system. Theoutput from this process is the continuous spectrumof errors at all spatial frequencies on the part. Eachcomponent in this block diagram is described below.

We are performing experimentation and valida-tion of each step in the error budgeting procedure ona T-based lathe. Test results are also discussed inthe following sections.

Transforming Errors into Spatial-Frequency Domain

Conventionally, a single peak-to-valley number isused to quantify the dimensional changes, not identi-fying the spatial frequency content of the error. Thisnew approach requires us to determine the fullfrequency spectrum of the errors. To do this, wemust relate the error characteristics to physicalproperties of the system. Forces generate thedimensional errors. The machine structure respondsto these forces due to the compliance of themachine, as shown in the block diagram of Fig. 2.

Often the forces are related to physical propertiesof the machine, such as the cycling of the rollingelements in bearing systems or the number of polesin a motor. However, when these forces are at ornear the machine resonances, the displacementscaused by these forces are amplified.

While the frequency content of the errorforces is often fixed in the spatial-frequencydomain, the machine resonance is fixed in thetemporal- frequency domain. The spatialfrequency is converted to the temporal frequencyby multiplying the spatial frequency by the veloc-ity. While the spatial frequency content of theforce error may be independent of velocity, the

spatial frequency content of the displacementerror is dependent on velocity.

For example, consider a machine with a reso-nance at 100 Hz. If the axis velocity is 10 in./min,then 600 cycles/in. is equal to 100 Hz in the temporal-frequency domain and is amplified by the machineresonance. However, if the axis velocity is100 in./min, then 60 cycles/in. is equal to 100 Hzand is amplified by the machine resonance.Therefore, the spatial-frequency content of thedisplacement error is dependent on the velocity ofthe moving components.

We observed this effect when we measured axialand radial error motions of the spindle on our testmachine. As expected, the air-bearing spindle is veryrepeatable with sub-micrometer levels of asynchro-nous motion. However, the error characteristicsdrastically change at different spindle speeds. Forexample, the axial motion at 840 RPM spindle speedhas a synchronous error with a dominant lobing of17, 18 and 19 cycles/revolution, as shown in Fig. 3.If the errors were associated with a physical prop-erty of the motor such as the number of commuta-tions, we would expect the spatial frequency of thelobing to remain fixed. However, at a spindle speedof 300 RPM, we observed a much higher spatialfrequency lobing pattern, as seen in Fig. 4.

In general, the spatial frequency of the lobingincreases with decreasing spindle speed. However,the temporal frequency of the dominant errorsremain in the same region for all spindle speeds, asshown in the plots on the right sides of Figs. 3 and 4.

We are investigating the source of the forcingfunction. It is curious that the forcing functionremains almost completely synchronous. Our hypoth-esis is that the forcing function is due the spindlespeed variations about the set point. This will beinvestigated further during the next fiscal year.

FY 98 3-5

Cuttingmodel ms2+bs+k

e-τs

1 X

CommandErrors

Fx

Fz

Machinedynamics

Regenerative displacements

Time delay (s)

Z

Figure 2. Block diagram of the machining process.



We also observed a similar effect with thestraightness errors. The cycling of the balls in thebearing system causes the lower frequencydisplacement errors. This component is fixed in thespatial-frequency domain and remains constant atdifferent feed rates of the axis. However, the higherfrequency displacement errors fell at the machineresonances, which are fixed in the temporal-frequency domain. Therefore, the spatial-frequencyspectrum of the displacement errors is dependenton the axis feed rate.

Combinatorial Rule

We have developed a combinatorial rule for theaddition of the frequency content of each elementalerror. The key to the combinatorial algorithm is toconsider the spectrum of each elemental error asthe sum of sinusoidal errors at specific frequencies.The addition of two sinusoidal signals at a givenfrequency results in a sinusoidal signal with thesame frequency, but the amplitude can varyanywhere from the direct difference to the sum of

the two amplitudes, depending on the phase shiftbetween the two signals.

We first identify all elemental errors that arecorrelated, and appropriately sum the amplitudes ofthese errors. We then consider the phase shiftbetween the remaining elemental errors to beuniformly distributed variables between 0 and 2π.We have analytically shown that the expected valueof the square of the net amplitude is equal to thesum of the squares of the amplitudes of eachelemental error. This is equivalent to saying that theexpected net power spectral density (PSD) is thesum of the elemental PSDs.

Furthermore, we can now determine the prob-ability distribution function of the net error withthe use of a Monte Carlo simulation. The 95%confidence limit of the net PSD is approximatelythree times the mean, and the 99% confidencelimit is approximately 4.6 times the mean. Thisis significantly less than the worst case error.For example, if 25 errors of equal amplitudewere summed, the worst case net PSD would beover eight times larger than the 95% confidence


180

90

60

30

0

330

300

270

240

210

150

120

0.46152

500400300200100035302520151050

Cycles/revolution (Hz)

0.03

0.06

0.05

0.04

0.02

0.01

0

Axi

al (

µm)

0.11538

0.34614

0.23076

180

90

60

30

0

330

300

270

240

210

150

120

0.15527

5004003002001000100806040200

Cycles/revolution (Hz)

0.010

0.025

0.020

0.015

0.005

0

Axi

al (

µm)

0.20502

0.10251

0.051255

Figure 3. Axial errorsat 840 RPM.

Figure 4. Axial errorsat 300 RPM.



limit, and over five times larger than the 99%confidence limit.

Material Removal Transfer Function

The purpose of the material removal transferfunction is to convert the motion of the tool in freespace to the motion of the tool in the part during thecutting process. This step is necessary becausecurrent error characterization procedures measurethe error motion of the tool in an open loop sense.The loop is closed when tool is in contact with thepart during the cutting process. Differences occurwhen the loop is closed due to static and dynamicstiffness of the machining process.

The conventional error budgeting procedureassumed that the measured motion of the tool infree space is the same as the motion of the tool inthe part during cutting. In other words, it assumesthat the transfer function equals one. We haveanalytically shown that the material removal trans-fer function equals one under the following assump-tions: 1) we have made multiple cutting passes onthe part; 2) the material removal transfer function islinear; and 3) the errors can be adequately repre-sented in the frequency domain with negligiblerandom components.

While the first assumption is valid, the secondand third assumptions are invalid. However, to afirst order approximation, we have experimentallydetermined that the material removal transferfunction is approximately linear around small devi-ations in the operating point. Furthermore, preci-sion machines often have very repeatable errorcharacteristics, so the third assumption is valid tofirst order.

Mapping the Errors into the Workpiece Coordinate System

Given the frequency content of the error motionof the tool during cutting, we must take intoconsideration that the path of the tool and the toolgeometry determine the frequency content of theresidual surface errors on the workpiece. Typicaltools with a round cutting edge impart a nominalsurface finish, or scalloping, during turning, evenfor a process with no errors. Next, we consider theexact path of the tool during the entire cuttingprocedure to map these errors onto the relevantworkpiece coordinate system.

For example, during a facing operation on adiamond turning machine, the part turns while thetool remains stationary. Consider the spatialfrequency content of a radial trace across the

workpiece. The turning process can be considereda sampling mechanism. The radial trace iscomposed of the time domain sampling of the toolmotion once every revolution of the part. Onceevery revolution, the tool falls on the radial trace ofinterest, leaving behind the signature of the tool aswell as any error motions.

The description of the process so far has been inthe time domain. However, we are interested in thefrequency domain. Sampling in the time domain canbe decomposed into a multiplication procedure ofthe original time-domain signal by a series ofimpulses. Since multiplication in the frequencydomain is equivalent to convolution in the frequencydomain, the sampling procedure is converted to thefrequency domain by a convolution process. Notethat unavoidable aliasing occurs for errors withhigher frequency content than the rotational speedof the spindle. Note also that errors at frequenciesthat are an even multiple of the spindle speed (suchas ‘synchronous’ spindle errors) do not appear onthe radial trace due to this aliasing.

The imparting of the tool geometry onto the work-piece can be considered a convolution in the timedomain. Conveniently, convolution in the timedomain is equivalent to multiplication in thefrequency domain. Therefore, the imparting of thetool geometry onto the workpiece in the frequencydomain can be considered a filter.

Future Work

Transformation of Errors into Spatial-Frequency Domain

During experimentation and validation, we havebeen able to measure the contributing errors.However, during the design process, we will nothave this luxury. Therefore, we must relate the phys-ical properties of the machine to the general types oferrors that are created. For example we discussedthe error motions of the spindle. We believe thatthese errors arise from the fact that the spindlespeed is varying, due to the spindle/motor/controllersystem. During FY-99 we will relate the physicalproperties of general machine components to thefrequency content of errors that are associated withthese types of systems.

Material Removal Transfer Function

For simplification we assumed that the materialremoval transfer function was linear. This is knownto be false. During FY-99 we will study the nonlin-earities associated with the cutting process and

FY 98 3-7



develop strategies to deal with these nonlinearitiesin the frequency domain.

Error Budget Procedure

At the end of FY-99 we will deliver an errorbudgeting procedure to predict the spatial-frequency content of errors on a machined part fora variety of machining conditions. The user willinput specifics about the machine components,structure, and control systems along with machin-ing parameters such as spindle speed and axisvelocity. Software will perform the appropriatecombinatorial algorithm and processing to predictthe spatial-frequency content of the errors on themachined surface of the part. With this tool, theuser can study the effect of changing machiningparameters or system components on the spatial-frequency content of the errors on the machinedpart. For example, the user will be able to replacethe spindle type or axis velocity and observe theeffects on the spatial-frequency content of theerrors on the machined part.

References

1. Evans, C. (1989), Precision Engineering: AnEvolutionary View, Cranfield Press, Bedford, UK.

2. Bendat, J. S., and A. G. Piersol (1986), Random Data:Analysis and Measurement Procedures, 2nd ed.,Wiley-Interscience, New York, New York.

3. Sokolnikoff, I. S., and R. M. Redheffer (1966),Mathematics of Physics and Modern Engineering, 2nded., McGraw-Hill, San Francisco, California, p. 319.

4. Donaldson, R. R. (1980), “Error Budgets,” in MachineTool Accuracy, Vol. 5 of Technology of Machine Tools:A Survey of the State of the Art by the Machine ToolTask Force, R. Hocken, ed., Ch. 9.

5. Slocum, A. H. (1992), Precision Machine Design,Prentice Hall, Princeton, New Jersey.

6. Thompson, D. C. (1989), “The design of an ultra-precision CNC measuring machine,” 39th CIRPGeneral Assembly, Trondheim, Norway, August20–26.

7. Shen, Y. L., and N. A. Duffie (1993), “Comparison ofCombinatorial Rules for Machine Error Budgets,”Annals of the CIRP, Vol. 42 (1), pp. 619–621.



recision Grinding of Brittle Materials


Introduction

For LLNL to meet the increasing demands of itsprograms, it is crucial that we extend our expertisein precision grinding.

The need for the development of cost-effectiveprecision fabrication of brittle material componentsis driven by the increasing demand for high-performance components from LLNL’s largeprograms, such as Weapons and Lasers. High-performance brittle materials such as silicon,glasses and a wide variety of ceramics will play anever-increasing role in many of LLNL’s and DOE’smajor programs. This project focuses on the preci-sion grinding of BeO ceramic components to be usedas heatsinks for mounted electrical components(Fig. 1). BeO is the preferred material for this appli-cation and others because of its rare combination ofhigh thermal conductivity (~56% that of copper) andits low electrical conductivity. The challenge isdeveloping a process to machine necessary heatsinkfeatures in BeO substrates while meeting optically-driven tolerance specifications.

Along with the need to develop precision fabrica-tion and process techniques, computer modeling andother analytical capabilities are instrumental astools to predict grinding wheel wear rates, materialgrindability and resultant characteristics of theworkpiece. In addition, metrology processes are

required to provide process information feedbackand to ensure the workpieces meet specifications.

LLNL’s precision grinding core technology devel-opment effort encompasses a number of relatedtasks, all of which play key roles in advancing preci-sion machining of brittle materials for programmaticapplications and give it great potential for success-ful commercialization with outside vendors.

Progress

Precision grinding of brittle materials encom-passes a variety of processes, including profile

FY 98 3-9

High performance brittle materials, such as silicon, beryllium-oxide (BeO) and glasses, offer high-performance properties for demanding engineering applications. Similar to the need that motivatedthe development of diamond turning capabilities at Lawrence Livermore National Laboratory (LLNL),the demand for precision-machined brittle material components is driving the development of preci-sion grinding. Precision grinding is often the only viable process to fabricate precision components ina cost-effective manner. The goal of our development project is to meet the needs of LLNL’s programsfor brittle material components that are difficult to manufacture. We focus on the process develop-ment and associated activities, such as process modeling, metrology and commercialization formedium- to high-volume production.

Mark A. Piscotty, Kenneth L. Blaedel, Pete J. Davis, and Pete C. DupuyManufacturing and Materials Engineering DivisionMechanical Engineering

Figure 1. BeO ceramic heatsink component.

315 Piscotty_qk 7/22/99 5:41 PM Page 3-9


grinding, cylindrical grinding, and surface grinding.1

In this project, the focus is profile grinding of intri-cate geometries in 2-mm-thick ceramic substrates.Figure 2a shows a schematic and dimensions of theexperimental components used in this study.2 Typicaltolerances required for this component range from±1 µm to ±5 µm. An SEM end view of a precisionground sample feature is shown in Fig. 2b. Note thatthe internal corners display finite radii, which is anindication of corner wheel wear. Wheel wear is amajor complicating factor in grinding small featuressuch as these, since a small amount of wheel wearcan result in out-of-specification workpieces.

In addition to the dimensions shown in Fig. 2a,other characteristics of the ground specimen alsohave stringent requirements. Figures 3a and 3bshow additional SEMs of a typical precision groundworkpiece. The long vertical wall shown in Fig. 3ahas both flatness and straightness tolerances of±1 µm along the groove length. Meeting theserequirements entails stringent control of the sidewheel wear and the ability to maintain the wheel ina free-cutting state. These two conditions are oftenadversarial because free-cutting wheels typicallyshed used diamond abrasives to expose sharp, freshabrasives, which itself is a form of wheel wear.Excessive side wheel wear can produce canted verti-cal walls, resulting in unusable workpieces.

Brittle materials are highly susceptible to edgechipping during processes such as precision grind-ing. Zero-tolerance edge chipping is typicallyrequired as it can degrade the strength and perfor-mance of the component. Figure 3b shows an SEMused to examine the edges of a feature bottom for


1.00 mm

0.150 mm

0.075 mm

BeO Ceramic

0.50 mm(a) (b)1.00 mm

(a)

(b)

Figure 3. (a) SEM of heatsink features; and (b) SEM of featurebottom notch.

Figure 2. (a) Typicalheatsink dimensions;and (b) heatsinkgroove end view.



edge chipping. The nominal grain size of the BeOused in this study is 15 to 25 µm. To generatesharp corner surfaces without edge chippingrequires that intragranular grinding take place.This requires that the grinding wheel maintain awell-dressed condition throughout the grindingcycle, ensured by intermittent dressing of thewheel during the grinding process.

A number of viable processes, each with itsadvantages and disadvantages, were possible candi-dates for machining these components. The processused at LLNL was selected because of its flexibility,robustness, and potential to be commercialized.Shown in the schematic in Fig. 4 are two separateapplications using the same basic process, one forcreep feed grinding of flat substrates (BeOheatsinks) and the other cylindrical grinding(ceramic engine components).

An on-line electrical discharge machining (EDM)system is used to impart precision profiles on ametal bond, diamond abrasive grinding wheel.Features on the rotating graphite EDM electrode areturned on its outside diameter surface using a singlepoint carbide tool and are used in the process tomachine the profiles on the grinding wheel. In thiscase, a grinding wheel with two profiles is optimal,since two grinding passes are required to complete agroove. However, this process can be extended togenerate additional profiles on a single grinding

wheel or to fabricate several profiled grindingwheels on a multiple wheel arbor. The ceramicworkpieces are held in a chucking fixture below theprofiled grinding wheel (Fig. 5).

Metrology

The unique heatsink configuration used in thisproject has several critical dimensions and requiresgeometric verification using off-line inspectionprocedures. These procedures involve a combinationof both visual and contact metrology. Visual inspec-tion is used for specific profile portions, such as the

FY 98 3-11

Figure 4. Schematic of machine tool set-up.

Grindingwheel

BeO workpieces

Figure 5. BeO heatsinks in grinding position.



flatness of the groove bottoms, the widths anddepths of the relief notch at the bottom, and the flat-ness of the groove walls. Contact metrology isperformed using a delicate stylus probe to allowinterrogation of the grooves’ inner geometries (wallsand bottom).

Figures 6a and 6b show inspection photographsof the process using a precision coordinate measur-ing machine (CMM) located at Sandia NationalLaboratory in California. The touch probe is held ina low force sensor head, and axes-positioning feed-back is provided from distance measuring laserinterferometers. This inspection procedure enableseach part to undergo 100% inspection, which is avaluable diagnostic during process development.As processes become more robust and proven,inspection procedures may be streamlined toincrease throughput.

Commercialization

While the primary goal of this project is todevelop precision grinding, a secondary goal is toenable this technology to be commercialized with atleast one outside vendor for higher volume andlower cost production for LLNL’s programs.

A number of challenges were encounteredduring the effort to commercialize this technology.Among these challenges were determining techni-cally competent vendors with the necessarymachine tools for precision fabrication of ceramiccomponents. In addition, the vendors must be will-ing and qualified to machine BeO, which is consid-ered toxic in its powdered state. This narrowed theselection down to one vendor, Brush Wellman inTucson, AZ. Brush Wellman is the sole supplier ofBeO in the U.S. and has significant experiencemachining BeO for customers.

However, the precision needed to fabricate theseheatsink components was beyond their experienceand we therefore are working closely with them totransfer our process technology to them. BecauseBrush Wellman’s machine tools and machining capa-bilities vary significantly from those at LLNL, wetailor the technology transfer to accommodate this.

Brush Wellman owns two machine tools that canmeet the stringent performance criteria to fabricateBeO heatsinks in medium lot sizes (about 100pieces per lot). The first is a surface grinder thathas been used as a workhorse for other precision-ground BeO components. The second machine tool,which we feel has the most promise for deliveringthese heatsinks at the lowest cost, is an MTI612precision slicing machine. This machine is capableof using a ganged wheel arbor (multiple wheelsmounted on one precision arbor) and has three-axes-positioning accuracy of better than 1 µm.Collaborations between LLNL and Brush Wellmanare establishing processes on both these machinetools at Brush Wellman, now the vendor of choicefor these heatsinks.

Modeling

The modeling development of this project centerson understanding the mechanisms for generatingand propagating sub-surface damage (SSD) duringthe precision grinding of brittle materials. This hasbeen studied by a number of researchers using avariety of models and techniques.3,4,5 The modelingtechnique investigated involves a continuum damagemechanics (CDM) model developed at the Universityof Connecticut under Professor B. Zhang and Ph.D.


(a)

(b)

Tactile probe

Tactile probe

BeO workpiece

Vision system

BeO heatsink

Figure 6. (a) Moore M48 CMM for heatsink metrology; and(b) set-up for tactile probe metrology.



candidate R. Monahan. The current CDM modelinvestigates the resulting SSD generated by a singlegrit material removal.

While this model has interesting implications andinformation, a new model will include materialremoval interaction among several grits in a grind-ing wheel, which is more characteristic of a realgrinding process. The CDM model introduces inelas-ticity and damage to accommodate the non-linearresponses of brittle materials. The current states ofeffective stress and damage and hydrostatic stressare used to simulate the cumulative anisotropicdamage of the brittle material. The next generationof this model will be developed at the University ofConnecticut with input and review from LLNL.

Conclusions

Precision fabrication of programmatically impor-tant brittle material components is a vital capabilitythat is being maintained and extended at LLNL.Precision grinding of brittle materials is often theprocedure of choice for this type of fabrication, sinceit offers many advantages over other possible meth-ods. This project leveraged a number of these advan-tages including process flexibility, readily availableprecision grinding machine tools and components,beneficial ties to industrial processes and vendors,and the ability to transfer a precision process to avendor for large volume commercialization.

Future Work

Clearly, grinding has a long history and recentdevelopments in the area of precision grinding areproducing remarkable results. However, precisiongrinding of brittle materials remains a relativelyyoung technology area compared to other tech-niques such as diamond turning, and thus cries outfor more research.

Future work should be focused on understandingthe fundamentals of the material removal mecha-nisms, wheel wear phenomenon, the dynamics of theprecision grinding process, and the propagation ofsurface and subsurface damage in the workpieces.Understanding the implications of how these mecha-nisms affect the precision of ground brittle materialcomponents is key to realizing the full potential ofthis fabrication process.

Semiconductor materials, thin films and opticalsystems are in the forefront of advanced materialsthat play a critical role in many LLNL programs.

New coating and fabrication techniques are produc-ing materials to meet the ever-increasingly stringentdimensional, defect and SSD requirements. To usethese materials to their maximum potential, theirdimensions and defect state must be measured withheretofore-unattainable precision.

Conventional SSD measurement techniquestypically focus on destructive methods, includingtaper polishing and tunneling electronmicroscopy.6,7 X-ray diffraction has been usedwith little success since it works best with well-defined crystalline substrates.

It became obvious during the course of thisproject that the technology to quantitatively evaluateSSD in a nondestructive manner demands substan-tial research and development. An in-situ, nonde-structive evaluation technique would be anextremely valuable and unique tool for interrogatingSSD as a result of grinding, lapping and polishing.

Current methods to quantitatively evaluate SSDare labor- and time-intensive, and destroy thesurface of the workpiece being evaluated. Drs. S.Soares (California Institute of Technology) and B.Zhang have developed a proposal for a nondestruc-tive method to evaluate SSD in brittle materials.

References

1. Komanduri, R., D. A. Lucca, and Y. Tani (1997),“Technological Advances in Fine Abrasive Processes,”Annals of CIRP, 46, pp. 1-52.

2. Drawing No. AAA98-107146-OA, LawrenceLivermore National Laboratory, Livermore,California.

3. Fahrenthold, E. P. (1991), “A Continuum DamageModel for Fracture of Brittle Solids Under DynamicLoading,” Journal of Applied Mechanics, 58(4),pp. 904-909.

4. Hu, K. X., and A. Chandra (1993), “A FractureMechanics Approach to Modeling StrengthDegradation in Ceramic Grinding Process,” Journal ofEngineering for Industry, 115(1), pp. 73-84.

5. Ju, J. W., and X. Lee (1991), “MicromechanicalDamage Models for Brittle Solids, I: TensileLoadings,” Journal of Engineering Mechanics,117(7), pp. 1495-1536.

6. Zhang, B. (1995), “Subsurface Evaluation of GroundCeramics,” Annals of CIRP, 44, pp. 263-266.

7. Xu, H. K. (1996), “Material Removal and DamageFormation Mechanisms in Grinding Silicon Nitride,”Journal of Material Research, 11(7).

FY 98 3-13


Center for Computational Engineering

propagation codes resulted in a 1997 R&D 100award for the code MELD. This work was continuedin the optical-full-wave algorithms being evaluatedand developed for use in target design codes.

In this report for FY-98, the computationalmechanics articles cover five code developmentactivities that have expanded our ability to handlelarge problem sizes on parallel computers andproblems that deal with greater physical complex-ity than before. The parallel code work is summa-rized in a single article covering all of computa-tional mechanics.

Two articles describe how the explicit DYNA3Dcode and the implicit codes NIKE3D and TOPAZ3Dhave been linked and coupled together.

For the code linkage, special element formula-tions were created to enhance the compatibility ofthe differing finite-element methods. Another arti-cle presents a substantial new cyclic viscoplasticconstitutive model added to our codes. The finalmechanics article talks about coupling mass diffu-sion and heat transfer.

The computational electronics and electromag-netics articles describe on-going code development,parallel implementation, design, and validationactivities for time-dependent and multi-lengthscale electromagnetics.

Articles summarizing physics modeling andcontrol of charged-particle beam devices for iner-tial fusion energy, microwave structure analysis,and accelerator design for LLNL’s advanced radi-ography mission are also in this section. Anotherarticle highlights this year’s development of newdiagnostics which have been developed for novellaser-electron interaction experiments aimed atdeveloping the next generation of light sources. Afamily of intercepting high-current electron beamdiagnostics was also developed. A final articlehighlights progress in various facets of computa-tional electronics, including electromagnetic andnuclear effects modeling.

The Computational Engineering Center is a vitaland growing component of the EngineeringDirectorate of Lawrence Livermore NationalLaboratory (LLNL). This new center is the result ofcombining the Computational Electronics andElectromagnetics Thrust Area with theComputational Mechanics Thrust Area.

The combined entity fuses the computationalexpertise of two organizations and represents aconsolidation of capabilities and activities. Thenumber of engineering analysts at LLNL has grownthis past year, as has the demand for sophisticatedscientific numerical simulations. The purpose of theComputational Engineering Center is to anticipateand provide for the future analytic needs of theEngineering Directorate at LLNL.

Activities and code development work in LLNL’sAccelerated Strategic Computing Initiative (ASCI)began this year with new efforts to parallelizeimplicit finite element methods. This long rangeeffort will bring the advantages of scaleablecomputing on massively parallel supercomputersto the entire suite of computational mechanicscodes (NIKE, TOPAZ, and DYNA) for importantprogrammatic applications.

Interactions with the Department of Defense(DoD) High-Performance Computing andModernization Program and the Defense SpecialWeapons Agency are of special importance as theysupport our ParaDyn project in its development ofnew parallel capabilities for DYNA3D. Working withDoD customers has been invaluable in driving thistechnology in directions mutually beneficial to theDepartment of Energy.

The research and development activities withinthe computational electronics and electromagneticsarea have yielded a set of simulation and designcodes that have contributed to three LLNL “R&D100” awards in the past two years: theory anddesign support for high-gradient insulator develop-ment, and antenna synthesis for microwave-basedbridge deck inspection, were 1997 and 1998 R&D100 application award winners; work in high-frequency electromagnetics modeling and optical

Peter J. Raboin and Clifford C. Shang, Center Leaders


Center for ComputationalEngineering

4


Contents

4. Center for Computational Engineering

OverviewPeter J. Raboin and Clfiford C. Shang, Center Leaders

Hybrid Ray/Wave Methods for Optical PropagationRichard P. Ratowsky, Jeffrey S. Kallman, Michael D. Feit, and Bedros B. Afeyan........................................4-1

Technologies for Advanced Induction AcceleratorsMaurice A. Hernandez ...............................................................................................................................4-7

TIGER: An Object-Oriented Time-Domain Simulation CodeDavid J. Steich, Jeffrey S. Kallman, Gerald J. Burke, S Terry Brugger, and Daniel A. White.......................4-9

OPUS: An Optically Parallel Ultrasound SensorJeffrey S. Kallman, Dino R. Ciarlo, Elaine Ashby, and Graham H. Thomas ...............................................4-13

Optical Transition Radiation Diagnosis for Electron BeamsGregory P. Le Sage and Roger A. Richardson...........................................................................................4-19

Characterization of Electromagnetic Scattering from Defects in the EUVL ProcessLisa Wang, Scott D. Nelson, Jeffrey E. Mast, and Abbie L. Warrick..........................................................4-23

Nuclear and Electromagnetic Radiation Simulation Tools for Dual-Revalidation of the StockpileDavid J. Mayhall and Michael F. Bland.....................................................................................................4-29

Self-Effects in Expanding Electron Beam PlasmasManuel Garcia .........................................................................................................................................4-33

Pump-Induced Wavefront Distortion in Prototypical NIF and LMJ AmplifiersMark D. Rotter, Kenneth S. Jancaitis, Christopher D. Marshall, Luis E. Zapata, Alvin C. Erlandson, Geoffroy LeTouze, and Stephane Seznec ...................................................................4-37

Parallel Algorithm Development for Computational MechanicsCarol G. Hoover, Robert M. Ferencz, Anthony J. De Groot, Robert J. Sherwood, Edward Zywicz, Yuen L. Lee, and Douglas E. Speck.................................................................................4-47

DYNA3D-TOPAZ3D Coupling and DYNA3D-NIKE3D LinkageJerry I. Lin ..............................................................................................................................................4-55

A Physically Stabilized Eight-Node Hexahedral ElementMichael A. Puso.......................................................................................................................................4-59

A Cyclic Viscoplastic Constitutive ModelPhani Kumar V. V. Nukala ........................................................................................................................4-65

Electromagnetic Cold-Test Characterization of the Quad-Driven Stripline KickerScott D. Nelson and James E. Dunlap......................................................................................................4-71

Photonic Doppler VelocimetryPaul D. Sargis, Nicole E. Molau, and Daren Sweider................................................................................4-77

Modeling Coupled Heat and Mass DiffusionArthur B. Shapiro and Philip M. Gresho...................................................................................................4-81

Analysis and Modeling of a Stripline Beam Kicker and SeptumBrian R. Poole, Lisa Wang, Yu Ju (Judy) Chen, and George J. Caporaso ..................................................4-85


ybrid Ray/Wave Methods for Optical Propagation


Introduction

Until the mid-19th century, natural philoso-phers argued vigorously over whether light was inreality a wave or a particle. Compelling reasonsexisted for both points of view: light seemed totravel mostly in straight lines like a particle, butshowed diffraction and interference like a wave.In the early 20th century, the same issuessurfaced again in the context of the quantumtheory, where both light and matter exhibited bothwave-like and particle-like behavior, depending onthe observational setting.

The ability of light to act like a wave or a particlehas practical consequences for the calculation ofits propagation. For systems large compared to awavelength, such as conventional bulk optics,particle-like geometrical optics is usually a goodapproximation. On the other hand, if the scale offeatures is on the order of a wavelength, such as ata focal point, solving a wave equation is necessaryto capture the physics of diffraction. Often, as in thebulk optics example, both circumstances arise insub-domains of the same problem.

Many practical situations important to LawrenceLivermore National Laboratory (LLNL) programs

share this dual length-scale property. One importantapplication is laser propagation in fusion plasmas.Traditional modeling uses ray tracing to transportthe laser intensity, and this is usually a good approx-imation. But when plasma gradients are wavelengthscale, as may occur in regions of NIF targets, a waveoptical treatment is essential.

Another relevant area is multi-mode photonics. Amulti-mode optical fiber may be optically large, butcalculating the effects of interference—speckle—between the modes is of critical importance.

The purpose of our proposal was to create acomputational tool which would move easilybetween the wave and ray optical regimes. Weaccomplished this by using phase space methods,where a set of rays distributed in a particular way inposition and angle retain many essential features ofwave propagation, including diffraction.1,2

By launching the right set of rays, diffractioncan be calculated directly from the ray distributionwithout explicitly solving a wave equation. In thisway, a problem domain can easily use both ray andwave optics in the regions where the descriptions aremost appropriate. To characterize and enhanceour understanding of the method, we developed aGUI-based photonics tool that can analyze light

FY 98 4-1

This aim of this project was to create a computational tool that bridges the gap between the waveand ray optical regimes, important for applications such as laser propagation in plasma and multi-mode photonics. We used phase space methods, where a set of rays distributed in a particular way inposition and angle retain many essential features of wave optics. To characterize and enhance ourunderstanding of the method, we developed a GUI-based photonics tool that can analyze light propa-gation in systems with a variety of axial and transverse refractive index distributions.

Richard P. Ratowsky and Jeffrey S. KallmanDefense Sciences Engineering DivisionElectronics Engineering

Michael D. FeitLaser Science and TechnologyLaser Programs

Bedros B. AfeyanPolymath Associates Pleasanton, California

TA405 Ratowsky_qk 7/22/99 5:48 PM Page 4-1


propagation in systems with a variety of axial andtransverse refractive index distributions.

This report gives an overview of our work inFY-98. First, we convey the principles of the phasespace method and give some simple examples. Then,we describe the software tool we developed to studythe propagation of fields through the Wigner method.This tool allows us to study the important issue ofthe appropriate sampling of rays in phase space toachieve desired accuracy.

Further examples of light propagation in systemswith a variety of axial and transverse refractiveindex distributions are given.

Finally, we describe some of the limitations of ourcurrent scheme and the path we envision for furtherdevelopment and applications.

Progress

Methodology

We usually think of an electric field as a vectorquantity that varies in space and time according toMaxwell’s electromagnetic field equations. We canalso describe the field in terms of the wavenumbersor spatial frequencies which comprise it (“Fourier

representation”). However, it is sometimes mostnatural to think in terms of a mixed representation,whereby the field is thought of as a set of spatialfrequencies, the spectrum of which changes withposition. This is analogous to the way music, whichis a pressure oscillation changing in time, can berepresented by a musical score which shows a setof pitches (frequencies) changing in time. For theelectric field, a wavenumber defines a direction inspace; the coordinates of the direction are a set ofangles. Thus, the field can be represented by afunction of position and angle, which defines the rayphase space.

One such function is known as the Wignerdistribution.3 Originally invented for quantummechanics, the Wigner distribution allows us tocalculate wave optical properties on the position-angle ray phase space.

Mathematically, the Wigner transform can bethought of as a Fourier transform not of the field,but of its correlation function relating the field attwo points in space. An example of a Wignerdistribution is shown in Fig. 1 for a Gaussian fieldand for a uniformly illuminated aperture.Qualitatively, the reason diffraction is included in theray description is that, at each point in space, one


Field

X

1 2-1-2

Vw

0.8

1.0

0.6

0.4

0.2

1.0

0.8

0.6

0.4

0.2

-1 -0.5 0.5 1

0

-2.5

2.5

-5

2

2

θ

θ

5

1

0

-1-2

(a)

(b)

W

W

X

1

2

0

1

1 -20

-10

0

1020

0

-1-0.5

0.5

0

X

X

Wigner transform

Figure 1. Wigner phase-space distributions for (a) Gaussian and (b) uniformly illuminated apertures.



has a fan of ray angles: this is just a manifestationof the Fourier uncertainty principle. The Wignerdistribution is not strictly a phase-space density,however, since it is not always positive.

The Wigner propagation algorithm works asfollows. From an initial complex electric fielddistribution, we calculate the Wigner transform.This distribution is then evolved in space bysimply transporting it along rays, that is, assum-ing conservation along a ray. This will be anexcellent approximation if the index is not toorapidly varying on a wavelength scale. If at everyposition in the ray phase space for the new (evolved)distribution, we integrate over all ray angles, weobtain the intensity distribution (“near field”) as afunction of spatial position.

Conversely, if at each angle in phase space weintegrate over position, we obtain the intensitydistribution as a function of ray angle (“far field”).Since in wave optics the near-field and far-fieldamplitudes are related by Fourier transform, thecomplex electric field can be reconstructed (up to aconstant phase factor) from the ray intensity distri-butions in position and angle.

As an example of a calculation using theWigner propagation algorithm, we calculated thediffraction pattern from a double slit (Fig. 2). TheWigner function is calculated in the plane of theslits, then propagated along rays to the plane of thescreen. The calculated result and the exact solutionare overlaid, and agreement is excellent. In fact, forfree propagation, the Wigner method is formallyexact, and the only errors are due to sampling. This

illustration shows strikingly that diffraction andinterference, usually considered outside the domainof ray optics, can be obtained through ray tracingthe Wigner distribution.

PHASTER

To study the Wigner method we developed aGUI-based code which allows us to propagate beamsin a variety of media in two spatial dimensions. Wecall the code “PHASTER”: Phase Space Techniquesfor Electromagnetics Research.

PHASTER allows us to set up an arbitraryinitial beam consisting of a sum of Gaussianswith selected widths and amplitudes. Aftercomputing and displaying the Wigner distributionfor the beam, it will solve the ray equations for a setof points in the ray phase space which sample theWigner distribution in a prescribed manner. The raysare traced using an adaptive Runge-Kutta methodthrough a variety of refractive index distributionshaving both transverse and axial variation.

After propagating the prescribed set of rays tothe exit plane, it will display the phase-space distri-bution at the exit plane, as well as the x-space andangle-space intensities (near field and far field).PHASTER gives useful insight into the dynamics ofthe rays in phase space and their effect on the waveoptical distribution.

Our first example of a PHASTER calculation isshown in Fig. 3. A Gaussian beam is shown propagat-ing through three “soft slabs,” that is, small regionswhere the refractive index rises and falls, depicted in

FY 98 4-3

X/λ

Intensity

Analytic intensity

Inte

nsi

ty (

a.u

.)

-20 -30 0

0.5

1.0

1.5

2.5

3.0

2.0

-10 0 10 20 30

20λ

200λ

5λ

Figure 2. Wignerphase-space methodcalculation ofdouble-slit diffractionby ray tracing.




Figure 3. PHASTERcalculation of propa-gation through threeslabs. The tilteddistribution on theright is the manifes-tation of diffractionin the ray phasespace.

Figure 4. PHASTERcalculation of wave-guide propagation,showing rays withtransverse turningpoints.



FY 98 4-5

Figure 6. PHASTERcalculation for speckled beam atnormal incidence oninhomogeneousplasma.

Figure 5. PHASTERcalculation foroblique incidence oninhomogeneousplasma with lineardensity gradient.



the central window of the GUI. The contour plot onthe left shows the initial Wigner distribution of theGaussian. Below it we see the x-space intensity, andto the left the angle-space intensity (far field). Thewindows on the far right show the analogous data forthe propagated distribution. Note that the phase-space distribution has become elongated and tilted:this is simply a graphic depiction of diffraction.

Rays at large-magnitude angles (top andbottom of distribution) move fastest in x, whilesmall-angle rays (near center of distribution)move little. The result is a spreading of the x-spacedistribution while the angular distribution does notchange, exactly what we would expect from simplefree-space diffraction. Due to the plates, additionalspreading occurs in x-space but not in angle,because rays receive no transverse impulse from theaxially varying index.

Figure 4 shows propagation in a waveguidingstructure. The rays are obviously confined to thewaveguide, and at the exit have formed a “galaxy-shaped” distribution. A given ray will encircle theorigin as it traces out a periodic path, but thefrequency of the rotation decreases as the angleincreases, causing the spiraling.

A third example is shown in Fig. 5. Here a beamis incident at an oblique angle on a linear indexgradient, a model for a laser propagating in aplasma. A subset of the rays is reflected by theplasma index gradient, while the rays gettingthrough form the distorted distribution on theright. This example is of particular interestbecause the problem admits an exact solution interms of Airy functions,4 and thus will allow directevaluation of accuracy.

Our final example (Fig. 6) shows the propagationof a speckled beam through the inhomogeneousplasma. By speckled we mean that we are launchinga set of Gaussians with a distribution of transversetilt angles and random phases. Propagation ofspeckle is important for laser-plasma interactionphysics and also for propagation in multi-mode fiber.Again, an exact solution to this problem is available.

An important feature of the code is its flexibilityin sampling the phase space at the entrance planeso as to put a higher density of rays where theWigner distribution has larger values. Sampling iscritical for the method, because if we do not samplewisely we suffer in computational efficiency for agiven accuracy. PHASTER allows us to divide thephase space into rectangular bins and vary the sizeof the subdivisions of the bins.

Future Work

The Wigner method shows great promise as awave propagation algorithm, and much remains tobe done to fully exploit its capabilities. To treatreflections accurately, for example, requiresextending the calculation to the time domain.

The method will be most effective when it isrelatively inexpensive to trace rays, as in bulk opticswhere all that is required is Snell’s law of refraction(and reflections if wanted) at boundaries. We arein the process of developing a 3-D bulk opticssolver. Up to now, we have ignored polarization,which can also be included. More subtle effects,such as accurately calculating tunneling, willrequire propagating complex rays.

We recognize that using rays is only a first stepfor the method to come to fruition. Because raysare 1-D, it is difficult to sample a 2- or 3-D spaceeffectively. One extension is to use “fat rays,” orGaussian basis functions, whose centers propagatealong a ray.

Ultimately, we envision using a wavelet basis,which can optimally sample a system that hasdisparate length scales. Resolution can be appliedonly where needed, and crude representationsretained where they cause minimal harm. This way,from rays to full wave optics in phase space, can besimulated by retaining a hierarchy of successivelymore accurate approximations.

In FY-99 we are focusing on applications of theWigner method to multi-mode photonics, whereeffective simulation tools are in demand but fullwave optics solvers are computationally difficult.5

References

1. Wolf, E. (1978), “Coherence and Radiometry,” J. Opt.Soc. Am. 68, pp. 6–17.

2. Bastiaans, M. (1997), “Application of the Wignerdistribution function in optics,” in The WignerDistribution: Theory and Applications to SignalProcessing, W. F. G. Mecklenbräuker, ed., North-Holland, Amsterdam.

3. Wigner, E. (1979), “On the quantum correctionfor thermodynamic equilibrium,” Phys. Rev. 40,pp. 749–752.

4. Afeyan, B. B. (1996), Bull. Am. Phys. Soc. 41, No. 7,paper 9Q2, p. 1598.

5. Ratowsky, R. P., B. B. Afeyan, J. S. Kallman, and M. D.Feit (1998), “Propagation modeling for multimodephotonics,” Lawrence Livermore National Laboratory,Livermore, California (UCRL-JC-131253-ABS).



echnologies for Advanced Induction Accelerators


Introduction

The Technologies for Advanced InductionAccelerators Project has just completed the secondyear of a three-year effort. This project is onecomponent of the Inertial Fusion Energy (IFE)research at Lawrence Livermore NationalLaboratory, aimed at developing the technologiesnecessary for a commercial fusion energy source.We have focused on the concept of indirect drivetargets using a heavy ion accelerator. A 1991 studyshowed that a recirculating accelerator (or recircu-lator) is a promising candidate for a cost-effectiveIFE driver. A recirculator is exactly what its nameimplies—an accelerator in which the beam travelsaround a ring-shaped configuration, repeatedlypassing through each accelerating element.

Progress

We have used the HIF Small Recirculator (whichpresently has one-quarter of the full ring completed)as a test bed. Our work in FY-98 concentrated on1) developing the components necessary for tailoredbeam acceleration and bending; 2) continued model-ing efforts aimed at constructing efficient beamsteering correction algorithms; and 3) conductingbeam dynamics experiments to further our under-standing of the beam control parameters.

Acceleration of the recirculating beam requireshigh-efficiency induction core materials driven byhigh-repetition-rate, programmable, pulse wave-forms. As the beam is accelerated, it gains bothkinetic and electrical energy. In addition, our accel-eration scheme calls for compressing the beam tohelp reduce emittance effects. The induction core

modulators must thus be able to produce precise,fast-rise-time waveforms with varying shapes,amplitudes, and widths.

During FY-98, we designed and built two proto-type versions of the required modulator. One versionis based on four parallel MOSFETs and associateddrivers configured as switching circuits, while theother version uses the MOSFETs as linear controlelements, complete with feedback and proportionaldrive circuitry.

The linear version has demonstrated acceptablevoltage regulation at average currents in excess of200 A, and peak currents in excess of 800 A. Thisperformance level is consistent with that requiredfor the recirculator. We expect to build and installmodulator boards (based on the linear prototype) todrive the induction cells on the existing 90º recircu-lator early in FY-99.

The dynamic beam parameters also require thatwe drive the beam-bending dipoles with varying volt-age levels. As the beam becomes more energetic,the dipole voltage necessary to provide the properbending increases. During FY-98, we installed andtested a prototype bending dipole pulser on thesmall recirculator. Pulser performance was mixed.Its power stages performed as expected, but weexperienced unexplained failures in some of theoutput transformers. In addition, we worked tomodify the pulser’s waveform generation and controlfeedback elements, to integrate the pulser into theexisting recirculator timing, control, and diagnosticssub-systems, and to compensate for the effects ofoutput loading and filtering.

We conducted beam dynamics experiments usingthe modified pulser, which determined the relativeeffects of timing, voltage level, and output ripple on

FY 98 4-7

The Technologies for Advanced Induction Accelerators project is aimed at developing the technologiesnecessary for a commercial fusion energy source. In FY-98, we developed components for tailored beamacceleration and bending, continued our modeling efforts, and conducted beam dynamics experiments.

Maurice A. Hernandez Defense Sciences Engineering DivisionElectronics Engineering

TA410 Hernandez_qk 7/22/99 5:53 PM Page 4-7


beam bending. We will conduct further bendingexperiments once the induction core modulatorshave been brought online.

We have used beam modeling throughout thisproject to develop steering control algorithms andto identify critical beam control issues. DuringFY-98, we studied the relative merits of two funda-mental steering methods, which will help determinethe best configuration for steering modules on thesmall recirculator.

The first method applies steering correctionsbased on measured values of beam position, veloc-ity, and momentum. This method tends to impartlarger “kicks” to the beam, using relatively fewersteering modules. The second method is based onsolving simultaneous equations designed to mini-mize functions based on measured beam displace-ments and steering module voltages. This methodtends to apply smaller “kicks” using steeringmodules in more locations. We will carry out experi-ments in this area during FY-99.


TA410 Hernandez_qk 7/22/99 5:53 PM Page 4-8

IGER: An Object-Oriented Time-Domain Simulation Code


Introduction

There is a significant need to advance our time-domain simulation capabilities. Projects desiringmore advanced CEM include antenna applications,vulnerability and susceptibility assessments,field/radiation effects due to outside stimulus,full-wave field behavior of kickers, splitters,induction cells, broadband phased arrays,Eigenvalues/Eigenvectors of lossy cavities, andlightning assessments, to name a few.

Time-domain CEM techniques have been in exis-tence for more than 30 years and have grown inpopularity. There are now hundreds of publicationseach year in the area as the techniques havematured. With so much previous work, it remains achallenge to keep up with the latest approaches.Often today it does not suffice to apply a singlemethod to a complex application. Although we areliving in an age where simplicity is preferred overcomplexity, the advantages of hybridized solutionscannot be ignored. However, the management of thiscomplexity comes at a price. This price is usuallydelayed technology advances and increased effort tosolve the problem.

The TIGER project endeavors to circumvent muchof the complexity involved in judiciously applyingmultiple algorithms to a given geometry to improveaccuracy and lower computational overhead. TIGERaccomplishes this management of complexity byusing object-oriented techniques.

This report describes our continuing progress inmodeling components for the Advanced HydrotestFacility (AHF) at Lawrence Livermore NationalLaboratory (LLNL), our parallelization progress,and our current efforts to model thin wires in thetime domain.

Progress

Last year we built a prototype version of TIGERand modeled a Beam Position Monitor (BPM), whichis essentially a 1/10-scale version (1/100 volume) ofa simplified kicker. We achieved a greater than100-fold reduction in the number of cells required tosimulate the BPM with comparable or improvedaccuracy. We also performed extensive testing andvalidation of the code.

FY 98 4-9

This report discusses our progress to date and our future plans for the Time-Domain GeneralizedExcitation and Response (TIGER) code. TIGER is an object-oriented computational electromagnetics(CEM) field solver. A prototype version of TIGER was built last year. This year we extended the codeto model more realistic structures, and began building the next version. We also studied the incorpo-ration of thin-wire algorithms. Our FY-99 plans include the continued development of the parallel-hybridized version of TIGER and the building of a powerful GUI for the code.

David J. Steich, Jeffrey S. Kallman, and Gerald J. BurkeDefense Sciences Engineering DivisionElectronics Engineering

S Terry BruggerComputer Applications Sciences and EngineeringComputation

Daniel A. WhiteCenter for Applied Scientific ComputingComputation

TA415 Steich_qk 7/22/99 5:55 PM Page 4-9


Testing and validation included comparisons withanalytic solutions, convergence studies, and carefulstudy of known error sources often found in thesesimulations, such as absorbing boundary conditionsand accurate modeling of sources and feeds.

Last year’s results were limited to monopole anddipole wakefield calculations of a BPM. This year wechose to model a full scale 3-D kicker structure builtas a prototype design for LLNL’s AHF programs, andobtained wakefield results for modes m = 0 throughm = 4. Previous modeling was limited to m = 0 andm = 1 for a scaled BMP geometry only.

Figure 1 is a photograph of a prototype accelera-tor kicker for LLNL’s Experimental Test Accelerator(ETA-II). Shown in Fig. 2 is a view of a section of aCAD kicker model. Figures 3 and 4 depict the realand imaginary transverse quadrupole wake imped-ances of the kicker, respectively. Figures 5 and 6depict normalized sextupole (m = 3), and dipole(m = 1) transverse wake potentials for an m = 3source, respectively.

Results show that the dipole wakes caused by asextupole excitation are not insignificant and mustbe taken into account. This is in agreement withexperimental evidence of the kicker. Preliminary test-ing of the kicker shows a triangular-shaped beamhitting a foil for large input beam radius (Fig. 7),suggesting a large sextupole wakefield. Research intounderstanding this phenomenon is ongoing.

Also in FY-98, we focused on building the nextmajor parallel version of TIGER. This next version willbe even more capable, flexible, and extensible. Shownin Fig. 8 is a plot of speed-up versus number ofprocessors for TIGER’s preprocessor on the ASCI Bluemachine. The geometry is a 234,000-cell mesh that ispartitioned on 1, 2, 4, 8, 16, and 32 processors. Weare getting excellent speed-ups up to 32 processors.

The slight decrease in speedup performance forthe 16- and 32-processor runs is due to the extracommunication that is involved. Since we are split-ting up the same geometry on successively more andmore processors, the surface to volume ratio isincreasing. This means there is less relative work todo for a given amount of required communication. Ingeneral, we must communicate the entire outersurface cell layer(s) of the geometry to adjacentprocessors. For 32 processors, there are onlyroughly 7312 cells per processor.


Figure 1. Photograph of prototype ETA-II accelerator kicker.

Figure 2. CAD enlargement of feed transition region forprototype ETA-II accelerator kicker.

Rea

l(Z

⊥2)

(kΩ

/m2 )

Frequency (MHz)

Wakefield codeTL model

200 400 600

0

4

12

8

0

Imag

(Z⊥

2) (

kΩ/m

2 )

Frequency (MHz)


200 400 600-4

0

4

12

16

8

0

Figure 4. Imaginary transverse quadrupole wakefield imped-ance for prototype ETA-II accelerator kicker.

Figure 3. Real transverse quadrupole wakefield impedance forprototype ETA-II accelerator kicker.



FY 98 4-11

If we would instead increase the geometry so thatthe number of cells per processor remained aconstant, our speed-up would again start toapproach perfect speed-up performance. For realproduction runs we would keep at least severalhundred thousand cells per processor. It is ourintent to eventually run problems with tens tohundreds of millions of cells.

Time-Domain Thin-Wire Capabilities

Adding thin-wire capabilities to TIGER wouldgreatly extend its usefulness to LLNL programs.There are applications in micro-impulse radar,nonproliferation, defense technologies, and theDepartment of Defense that could utilize thin-wiresimulation capabilities. While finite-difference andfinite-volume techniques are very powerful and havefound widespread use in electromagnetics, modeling

thin wires in such algorithms presents several chal-lenges. The wires are usually much thinner than thedesired cell size. Several methods, which use eithercontour path integration, integral equations for wirecurrents, or transmission line equations have beenused. Each of these methods has different advan-tages and disadvantages. However, each of theabove mentioned techniques requires the wires to liealong electric or magnetic mesh edges, whichrestricts the wire geometry when cells are orthogo-nal. Furthermore, with more general meshes,generation of a new mesh is required when awire is moved.

Our goals for FY-98 were to implement a thin-wirealgorithm and see if we could generalize the algo-rithm to move the wire laterally along the grid. Thiswould be the first step toward having an arbitrarywire within the grid without having to mesh the wire.

We successfully implemented thin-wire algo-rithms by Holland1, Riley2, and Taflove3. We alsoimproved the accuracy of Holland’s method andgeneralized the algorithm to allow for wires movedlaterally within a mesh. Results for our new schemeare shown in Fig. 9 for the real part of a dipoleimpedance for a thin wire of 0.001-m radius with acell size of 0.035 m for various wire locations withinan FDTD cell.

Figure 6. Normalized dipole (m = 1) transverse wake potentialfor an m = 3 source.

Sext

upo

le w

ake

po

ten

tial

W⊥

3

s (m)2 4 6 8

0

0.4

1.2

0.8

0-2

Figure 5. Normalized sextupole (m = 3) transverse wakepotential for an m = 3 source.

Sext

upo

le w

ake

po

ten

tial

W⊥

3

s (m)2 4 6 8

0

0.4

1.2

0.8

0-2

Figure 7. Experimental results for triangular-shaped beamhitting foil, for large input beam radius in the prototype ETA-IIaccelerator kicker.

Figure 8. TIGER preprocessor speed-up factor vs number ofprocessors on ASCI-blue.

Spee

dup

fac

tor

Number of processors15 20 25 30

15

20

30

25

5 10

5

10

IdealActual



Shown in Fig. 10 is Holland’s scheme for thesame dipole impedance. Note the large variation inthe real part of the dipole impedance of Holland’sscheme for various wire locations.

Future Work

Next year, our technology-base focus will be toprovide a powerful graphical user interface (GUI) forTIGER. This GUI will be object-oriented and allowmany of TIGER’s capabilities to be used in a user-friendly environment. We will continue to extend theparallel version of TIGER. Extensions will include theparallelization of the solver and the simultaneoushybridization of mesh types and physics algorithms.

References

1. Holland, R. (1981), “Finite-Difference Analysis ofEMP Coupling to Thin Struts and Wires,” IEEE Trans.Electromag. Compat., Vol. EMC-23, No. 2, May.

2. Riley, D. J., and C. D. Turner (1998), “The VOLMAXTransient Electromagnetic Modeling System,Including Sub-Cell Slots and Wires on RandomNon-Orthogonal Cells,” Fourteenth Annual Reviewof Progress in Applied ComputationalElectromagnetics, March.

3. Umashanker, K. R., A. Taflove, and B. Beker(1987), “Calculation and Experimental Validationof Induced Currents on Coupled Wires in anArbitrary Shaped Cavity,” IEEE Trans. Antennasand Propagation, Vol. 35, pp. 1248–1257.


Re(

Z)

(Ω)

Frequency (MHz)200 400 600 800

-500

3000

2500

2000

1500

1000

500

0

0

∆x

b

da

c

Wire position

NECabcd

Figure 9. Thin-wire results for dipole moved to various locations within a cell.

Re(

Z)

(Ω)

Frequency (MHz)200 400 600 800

-300

2700

2400

18001500

1200

900

600

2100

0

0

300

NECabcd

Figure 10. Holland’s thin-wire results for dipole moved tovarious locations within a cell.


PUS: An Optically Parallel Ultrasound Sensor


Introduction

Ultrasound imaging is used to inspect parts,monitor processes, and diagnose problems in humanpatients. Soon after x-ray CT was invented, attemptswere made to do the same kind of imaging withultrasound. One of the many problems encounteredwas the difficulty of acquiring the necessary data ina timely fashion. Other problems included thetendency of ultrasound not to travel in straight lines,and to require a couplant between source, object,and sensor.

The existence of these problems is unfortunate,considering that 3-D volumetric imaging by ultra-sound would be quite useful for screening for breastcancer. Consider the advantages: no ionizing radia-tion, no compression, and no toxic developing solu-tions or film. At present the only way to get the infor-mation necessary to do this kind of imaging is to usearrays of piezoelectric sensors, and either multiplexthem in time or have multiple copies of the necessaryelectronics to read them, which is expensive. Thesensor we are developing addresses these problems.

Our new sensor uses the classical opticalphenomenon, frustrated total internal reflection(FTIR), to make the incident ultrasonic wave modu-late a beam of light, which is then acquired by a

camera and computer. A sequence of images, eachtaken at a different source acoustic phase, enablesus to reconstruct the ultrasonic phase and ampli-tude over an entire 2-D surface.

In this report we detail the development anddesign of our optically parallel ultrasound sensor.

Optics

FTIR is a consequence of the wave nature oflight.1 When light moves from a slow medium to afast one, there is a critical angle, θc = sin–1 (n2/n1)(where n1 is the index of refraction of the slowmedium and n2 is the index of the fast medium),beyond which the light is totally reflected. However,an evanescent wave extends a short distance(approximately one wavelength) into the fastmedium, and if another piece of slow medium inter-cepts this evanescent wave, some of the lighttunnels across the gap (Fig. 1). How much lighttunnels is a function of the angle of incidence, theindices of the media, and, most importantly from ourpoint of view, the size of the gap.

We can exploit this effect to build an array ofacoustic pixels. Each acoustic pixel is composed of athin silicon nitride membrane suspended on shortgold walls over an optical substrate. The gap

FY 98 4-13

This development project addresses the need for a faster, less expensive method of transmissionultrasound. It uses the principle of frustrated total internal reflection to transduce acoustic pressureinto optical modulation. An entire 2-D plane of data at a time can be acquired. We describe themodeling and verification of a final sensor design.

Jeffrey S. KallmanDefense Sciences Engineering DivisionElectronics Engineering

Dino R. CiarloEngineering Research DivisionElectronics Engineering

Elaine Ashby and Graham H. ThomasManufacturing and Materials Engineering DivisionMechanical Engineering

420 Kallman_qk 7/27/99 7:31 AM Page 4-13


between the membrane and the optical substrate isfilled with air (Fig. 2). The entire assembly isimmersed in the ultrasonic couplant. When an ultra-sonic pressure wave impinges on the acoustic pixel,the membrane deflects, causing a change in theamount of light reflected from the total internalreflection surface of the acoustic pixel. An entirearray of acoustic pixels can be used to modulate abeam of light.

The information we desire from this sensor is thephase and amplitude at each acoustic pixel. Toobtain this information we illuminate the sensorwith ten sequences of optical pulses, each sequencetimed to act as a strobe light at a specific acousticphase (Fig. 3). We extract the phase and amplitudeat each pixel by fitting the intensity at that pixelthrough the sequence to the form I = B + Asin(2πi/10 + φ), where I is the intensity, B is thebackground, A is the amplitude of the sinusoidalvariation, i is the index of the image in the sequence,and φ is the phase of the variation.

The optical train of this device is as follows: weilluminate the sensor using an LED, the light fromwhich is homogenized, polarized, and collimated. Weacquire the reflection from the sensor using a CCDstill camera.

Acoustics

The inspection technique we are interested in istransmission ultrasound. In this modality, anacoustic source sends out pressure waves through acouplant, such as water, oil, or medical ultrasoundgel, to the object of interest. The pressure waves aretransmitted through the object, being changed inamplitude and phase along the way. The pressurewave emerges from the object of interest and travels,via the couplant, to an acoustic sensor. The sensordetermines the acoustic phase and amplitude, either

at a point, on a line, or in a plane. The sensor we aredeveloping will measure the acoustic phase andamplitude of an entire plane at a time.

Our sensor works because the pressure wavecauses the flexing of a membrane, causing it tovibrate with a phase and amplitude that are func-tions of that wave.

In designing our sensor, we needed to have amembrane with a frequency response high enoughto vibrate at the frequencies of interest to us(approximately 1 MHz).

Progress

Modeling

We began our work on this project by modeling asmany of the systems and processes as possible. Weused TSARLITE to model the optical elements of thesensor (FTIR), DYNA3D to model the acousticresponses of the membranes and their supports, and


Fastmedium

Slowmedium

Fast medium Gap

(a) (b)

θc

Slowmedium

Slowmedium

Figure 1. (a) Illustration of total internal reflection. Lighttrying to exit the slow medium at an angle higher than thecritical angle will be completely reflected. (b) Frustrated totalinternal reflection. The evanescent wave that extends into thefast medium overlaps another piece of slow medium, andsome of the light tunnels across the gap. The amount of lightthat tunnels through the gap is sensitively dependent on thesize of the gap.

Top view Side view

Uniformoptical

illumination (strobed)

Data-richoptical output

Wall

Membrane

Water

λoptical/4Air

Phase 1 Phase 2Optical pulse

Phase 3

50 ns 1 µs Acoustic wave

Figure 2. Illustration of an acoustic pixel, a thin membranesuspended over an optical substrate by gold walls. There is anair gap between the membrane and the optical substrate, andthe acoustic pixel is exposed to water. An ultrasonic pressurewave moving through the water will cause the membrane tovibrate, changing the air gap enough to modulate a beam oflight reflected from the total internal reflection surface of theoptical substrate.

Figure 3. Illustration of the illumination of the opticalsubstrate at particular times during the acoustic wave. Thiscauses the optical beam to act as a strobe light. A sequence often images allows us to deduce the motion of each acousticpixel through the acoustic wave, and thus the pressure andphase of the acoustic pixel excitation.



a modification of BEEMER to model the imagingsystem as a whole.

Using TSARLITE, we were able to determine theranges where we could expect FTIR to be useful, andbounded the permissible thickness of the membraneand the heights of the supports it would stand upon.This aspect of the modeling was performed to makecertain that we would be able to engineer to thephysical phenomenon we were using.

The modified BEEMER program was used to exam-ine the issues that will arise when we use the sensorfor diffraction tomographic imaging. It was used tomodel the tomographic data acquisition processes, aswell as a variety of reconstruction algorithms.

The simulation program we used most wasDYNA3D. We used this program to model our firstsensor design, a membrane suspended on an arrayof posts. DYNA3D showed us that this design hadneither the sensitivity nor the frequency responsenecessary to allow us to acquire the data we needed.Guided by our simulations, we developed a moreresponsive design, a set of resonant membranessuspended on walls (Fig. 2). Simulation showed thisdesign was responsive and sensitive, but had prob-lems with cross-talk and drift of the resonantfrequency as a function of hydrostatic pressure(Fig. 4). Further simulation allowed us to modify thedesign so as to greatly reduce cross-talk (Fig. 5).

Simulation eliminated the need to experimentblindly, but we still needed to do experiments toverify that the simulations were correct, and getproof of principle results. To do these experimentswe built a test system.

FY 98 4-15

Edge length = 20 µm




Res

on

ant

freq

uen

cy (

Hz)

x 1

06

Pressure (psi)

1.00

2.00

3.00

4.00

5.00

6.00

0.00 1.00 2.00

Materials 1

2x

y

z

(a) (b)

Figure 4. Graph showing how the resonant frequency of a setof simulated acoustic resonators varies with the hydrostaticpressure load.

Figure 5. Simulation Results. (a) All of the acoustic pixels in this simulation have the same resonant frequency. An excitation of therear left pixel spreads to all the surrounding acoustic pixels. (b) Staggering the sizes of the acoustic pixels greatly reduces the spread-ing of energy (cross-talk).



Verification

The test system we built consisted of the testmembrane, a tank with an acoustic source, and theoptics. The test membrane was designed to allow usto examine the responses of a wide range ofmembrane resonator sizes to variations in hydro-static pressure. The rest of the system was designedto make all of the parameters of interest easilyavailable for manipulation.

The test membrane consists of a 1-cm-by-1-cmmembrane of silicon nitride, supported by goldwalls, held in a silicon frame. The gold walls arepatterned as shown in Fig. 6. On the left hand sideof the test membrane are alternating rows of 60acoustic pixels ranging in size from 20 µm to 80 µmon a side. On the right hand side of the testmembrane only half of the rows are populated withacoustic pixels. The reason for the wide range insize is evident from Fig. 4.

We suspected that only acoustic pixels that wereat the exact resonant frequency would be sensitiveenough to modulate the optical input, and soattempted to include all pertinent sizes. In additionto the wide range in resonator sizes available wehad membrane fabrication parameters available formodification as well, that is, membrane thicknessand gold wall height.

The remainder of the system is illustrated inschematic form in Fig. 7. The oscillator provides a1-MHz signal, which is amplified and fed throughthe power meter to the acoustic source in the

water tank. The same signal goes to the pulsegenerator, which in turn excites the optical sourceas a strobe light. The pulse generator output andthe oscillator output are both fed to an oscillo-scope to allow the user to place the optical pulseat any point in the acoustic phase. The opticalpulses are homogenized, polarized, collimated, andfed through the prism to the sensing surface, andthe reflected light is captured by the camera andsaved in the computer. The user has control overthe oscillator frequency, the amplifier gain, thecamera exposure time, the optical pulse width,amplitude, and placement in the acoustic phase,and the depth of the water in the tank.

This is the procedure followed during a typicalexperimental run: the water tank is filled to thedesired depth, and the acoustic power level, cameraexposure time, optical pulse width, and optical pulseamplitude are set. Four sequences of ten images areacquired, each with the optical pulse occurring at adifferent acoustic phase.

We learned a great deal using the test system,most importantly that: 1) we can engineer to thesizes and tolerances necessary to use the physicalphenomenon; 2) we can extract phase and amplitudedata at each acoustic pixel from sequences ofimages; 3) the resonances are so broad as to makeacoustic pixel size non-critical; and 4) hydrostaticpressure causes little change in membrane response.

After performing numerous experimental runs wewere able to arrive at an optimal set of design para-meters for the final sensor.


1 cm

1 cm

Water tank

Acousticsource

Power meterAmplifierOscillator

Pulse gen.

Computer

Oscilloscope

Homogenizer

PolarizerLens

Sensor

Camera

LensLED

Figure 6. Schematic of the pattern of acoustic pixels in the testmembrane. On the left are alternating rows of 60 acousticpixels ranging in size from 20 µm to 80 µm on a side. On theright only half of the rows are populated with acoustic pixels.

Figure 7. The test system. The oscillator drives the amplifierand pulse generator with a 1-MHz sinusoid. The acousticsource insonifies the test membrane which is illuminated with50-ns optical pulses strobing at a user-specified acousticphase. Images of the membrane are acquired by the cameraand stored and processed in the computer.



Future Work

We have completed the design for the finalsensor. The final sensor is currently in the processof fabrication. Since we have funding to continuethis work in the next fiscal year, we will completethe construction of our final sensor, calibrate it,and use it to collect phase and amplitude datathrough a variety of test subjects. Initially we willonly make 2-D transmission ultrasound images, butultimately we will generate volumetric images of 3-Dobjects using data from this sensor to feed a diffrac-tion tomography code.

References

1. Yeh, P. (1988), Optical Waves in Layered Media, JohnWiley and Sons, New York, New York.

FY 98 4-17


ptical Transition Radiation Diagnosis for Electron Beams


Introduction

Transition radiation is produced when a chargedparticle traverses a barrier between two media withdifferent dielectric characteristics. This radiationemerges with a distinct angular distribution, whichcan be exploited to reveal electron beam character-istics that are otherwise difficult or cumbersome tomeasure. The basic configurations for thesemeasurements include a single or multiple samplesof dielectric or metallic material inserted into theelectron beam. Using appropriate materials andoptical measurement techniques, the spatial profileand the divergence of the electron beam can bemeasured at a single point, in a single shot.

Though the divergence of the beam is degradedby the thin (few micron) dielectric foils, it is notcompletely disrupted, as in the case of traditionalslit projection techniques used to measure the sameinformation. These seemingly delicate foil sampleshave in fact proven to be quite robust in some cases,accepting long macropulse lengths, high-averagecurrents, high-energy beams, and relatively tightfocal spots.

The equations describing this radiation are wellknown, but are somewhat cumbersome to use forreal materials, particularly for relativistic particles.Typically, approximations are made so the equations

can be used analytically. The materials used toproduce transition radiation are described bydielectric constants in the desired frequency range,both real and imaginary parts. Simplified forms ofthe transition radiation formulae were programmedfor comparison to measured data, and for estima-tion of flux levels at angles away from the peaks ofthe OTR pattern.

A characteristic of the radiation pattern at highenergies is the directivity of the back-scattered radi-ation. For a relativistic factor γ of 100, OTR isconfined to tens of milliradians of the reflectionangle. The reflection angle is equal to the incidentangle, so that for a foil angled at 45° with respect toan electron beam, the radiation pattern will becentered at 90°.

This is convenient for most applications, but forthe confined space available in the FXR experiment,it was not possible to view transition radiation at theproper angle of maximum intensity. To broaden theradiation pattern, the surface of an OTR foil wasroughened so that a more random incident angle ispresented to the electron beam.

The effects of surface roughness on an OTR radia-tion pattern were modeled using MATHEMATICA.The surface roughness was modeled as a randomGaussian function, centered at the foil angle. AnOTR distribution was integrated over this Gaussian

FY 98 4-19

We describe the accomplishments of two technology-base projects dealing with the development ofelectron beam diagnostics using optical transition radiation (OTR). Diagnostic experiments usingOTR were performed at the Lawrence Livermore National Laboratory (LLNL) flash x-ray facility(FXR), and at the LLNL 100-MeV RF linac facility. The original charter for the RF linac was the devel-opment of diagnostics appropriate for a high-peak-brightness electron beam. OTR appeared to be asuitable candidate for measuring micron-scale beam spots, and milliradian divergences. Transitionradiation diagnostics were evaluated using the thermionic electron beam produced by the LLNL RFlinac facility. The diagnostics are intended for use with the high-brightness electron beam that will beproduced by a photocathode injector currently under development. The FXR effort concentrated onthe use of OTR for transverse beam profile imaging at viewing angles away from the OTR radiationpeaks, at time scales short compared to the bunch length. The effects of surface roughness on theradiation pattern of OTR were examined theoretically.

Gregory P. Le Sage and Roger A. RichardsonDefense Sciences Engineering DivisionElectronics Engineering

425 LeSage_qk 7/27/99 7:33 AM Page 4-19


function to observe the effects of this roughnessmodel. When the foil surface roughness wasincreased in the model, the radiation lobes werebroadened. Light produced by OTR was successfullymeasured at the FXR using a roughed KAPTON foilfar from the optimum specular angle.

Progress

The efforts and accomplishments for the RF linacfacility for FY-98 are described in this section. Wehave expanded the scope of the effort to includepreparation of the photocathode electron accelerator.

A single foil OTR measurement in FY-98 providedthe first emittance data ever collected on the high-energy S-Band RF linac beam at LLNL. The tech-nique was proven for our beam parameters, andthe diagnostic will certainly be used as the machineis improved.

A theoretical study of the angular distribution oftransition radiation was performed to evaluate themeasured data under our particular conditions. Onecan imagine that while a single electron produces aperfect angular distribution of transition radiationlight, a broad thermal distribution of electrons witha large spot size produces a blurred pattern withcharacteristics that are not readily tractable. Asimulated distribution of electrons with imposedspatial and angular distribution (MATHCAD-scriptdeveloped) was used to generate characteristic OTRpatterns that could be compared to measured data.Added complication results from the large spot size,which affects the optical collection of the data, anddepends on the distance to the point of measure-ment (a CCD).

Taking all relevant effects into account, a multiple-parameter fit was performed to compare the theo-retical and measured distributions (C++-code devel-oped). Error bars were thus established between themeasured data and the theoretical fit over a range ofmeasured angles, and an emittance with specifiedtolerance was measured.

When two or more foils are used for a transitionradiation measurement, an interference pattern isproduced by the forward-projected OTR light fromthe first foil, and the specularly reflected light fromthe adjoining foils. This technique is useful sincemultiple peaks and nulls allow more precise evalua-tion of the beam divergence.

From a practical perspective, the angular patternof light coming from a multi-foil arrangement ismuch less ambiguous than that from a single foil,

since the interference fringes appear to sharpenconsiderably as the beam and optical system para-meters are properly adjusted.

Just such an effect was measured this year, againusing the LLNL linac beam, and a pair of aluminizedmylar foils. This measurement also provided anothersignificant milestone in the life of the linac beam.For the first time, we had visual feedback for tuningthe divergence of the linac beam at a given point.Starting with a beam spot optimized for small sizeand regular (round) appearance, we noticed that theOTR interference pattern was quite poor. As thefringe sharpness was optimized by tuning the linacoptics, we found that the spot image became muchmore irregular, indicating that the beam transportsystem was astigmatic (horizontal and vertical focallengths were not equal). With a remotely rotatablepolarizer, the collimation in orthogonal directionscan be investigated. These capabilities will provideessential information for interaction experimentsbetween the linac beam and a short laser pulse.

In addition, the prompt light produced by transi-tion radiation can be used to characterize thetemporal characteristics of an electron beam. TheFXR effort described previously included the use of agated intensifier to measure the beam evolution.Short pulse beams can be similarly characterizedusing a streak camera.

Measurement of the OTR light produced by thetwo-foil setup using a fast-risetime PMT wasperformed at the RF linac facility, confirming the RFsignals produced by low bandwidth beam currentmonitors. With the bandwidth of the PMT (3 GHz),we were able to detect changes in the macropulse asa function of time.

Future Work

In FY-99 an experiment is planned to relay-imageOTR light to an optical test bench, where the imagecan be masked to allow precise measurement ofdivergence characteristics across the beam profile. Ifa mask with multiple apertures is used, an “opticalpepper-pot” measurement will be possible, allowingreconstruction of the beam phase space. This tech-nique would be extremely useful for the high-energylinac beam, since a collimator for a 100- to 150-MeVelectron beam would be many centimeters thick. Theimaging arrangement is near completion, with relayoptics and wall penetrations in place. Though theconcept for this technique is patented, this diagnos-tic has never been implemented before. We are


425 LeSage_qk 7/27/99 7:33 AM Page 4-20


currently discussing this and other collaborativeexperiments with the two authors of the opticalpepper pot diagnostic patent.

OTR has been successfully demonstrated as auseful diagnostic for two very different classes of rela-tivistic electron beams. Closer ties are being forgedbetween the RF linac group, FXR, and ETA II. Furtherdevelopment of these diagnostics is expected.

FY 98 4-21

425 LeSage_qk 7/27/99 7:33 AM Page 4-21

haracterization of Electromagnetic Scattering from Defects in the EUVL Process


Introduction

In recent years, there has been an increasinginterest in extreme ultraviolet lithography (EUVL),which is a promising technology for producingsemiconductor features with resolutions <0.07 µm.The use of EUV technology would permit the devel-opment of computer chips with faster speed andmore memory storage. At EUV wavelengths, allmaterials become very absorptive. Therefore,unlike transmissive masks used in traditional opti-cal lithography, this technology uses multi-layercoated reflective masks.

Currently, the mask blanks are constructed bydepositing 81 alternating layers of molybdenum (Mo)and silicon (Si) on an Si substrate. Each layer pair is7 nm thick (3 nm of Mo and 4 nm of Si), producing a284-nm high multi-layer coating. The masks are thenmade by depositing absorber patterns on top of theMo/Si multi-layer mask blanks.

Figure 1a shows the complete EUV lithographicprocess. The EUV light, created using a laser plasmasource, is directed toward the mask using reflectiveMo/Si multi-layer coated condenser optics. After theEUV light passes through the mask containing thecircuit patterns, it then passes through the multi-layer imaging optics to reduce the image size toallow very small circuits to be printed.

The multi-layers are deposited using a newlydeveloped ion beam deposition system, IBD-350,shown in Fig. 1b. This system is capable of produc-ing uniform low-defect-density multi-layer-coatedreflective mask blanks (Fig. 2).1,2 In depositingdifferent layers of material for a reflective mask,ions collide with the target of Mo or Si, and are thenprecisely deposited on the substrate.

For EUVL, even small defects on the substrate orin the multi-layer mask blank may pose a significantproblem. It is not only because of the small featuresize, but also because very small defects may intro-duce compound errors from the multi-layer deposi-tion process, which can then cause problems whenthe circuit patterns are printed. Preliminary print-ability models suggest that defects >30 nm in diame-ter may print in the EUVL lithographic process.3

There is considerable concern about the effect ofdefects in the EUVL mask blanks caused by contami-nating particles, substrate scratches, and otherimperfections that may result in printable defects.The identification of printable defects is essential tothe development of projection EUVL. This is thesubject of an ongoing experimental program atLawrence Livermore National Laboratory.

The development of a realistic scattering model topredict how electromagnetic waves interact withanomalies in the multi-layer coating of the mask

FY 98 4-23

Extreme ultraviolet lithography (EUVL) is a promising technology for producing high-resolutionsemiconductor features in next generation lithography. For EUVL, even small defects can pose asignificant problem. Scattering models of defects, both for the substrate and the mask blank, areneeded to understand the critical issues associated with mask defects. In FY-98, we have appliedboth asymptotic/analytical methods and numerical techniques to model the electromagnetic scatter-ing from substrate defects to study how electromagnetic waves at optical or EUV wavelengths inter-act with anomalies on the substrate.

Lisa Wang and Scott D. NelsonDefense Sciences Engineering DivisionElectronics Engineering

Jeffrey E. Mast and Abbie L. WarrickLaser Engineering DivisionElectronics Engineering

430 Wang_qk 9/8/99 6:16 PM Page 4-23


blank is needed for the identification of printabledefects. In addition, the scattering cross-section ofprintable EUVL defects is needed for the develop-ment of inspection tools and effective diagnosticalgorithms at both optical and EUV wavelengths.

Progress

Both asymptotic/analytical methods and numericaltechniques were applied to model the electromagneticscattering from defects on the substrate. The applica-bility of asymptotic/analytical methods as a functionof defect size and shape was first evaluated.

Within their range of validity, asymptotic methodsincrease the computational efficiency at leasttenfold, compared to current numerical methods,helping us gain physical insights into scatteringmechanisms, and facilitating the interpretation ofnumerical solutions. Several asymptotic/analyticalmethods were examined, among which are the phys-ical optics approximation (which is applicable whenthe defect is large compared to a wavelength);4-7 thephysical theory of diffraction (which takes intoaccount the diffracted rays due to edges, corners,and curved surfaces);6,8 the shooting and bouncingray technique;8 and radiative transfer theory.7

Electromagnetic models were developed tosimulate the scattering from substrate defects andparticles including spheres, cylinders, disks, ellip-soids, and hemispheres. Given an incident wave Eiin the direction θi, φi, the first-order solution of theback-scattered electric field from a substratedefect can be expressed as

+ +( ) ( ) – , ; – ,f Rqp i i i i p iπ θ π φ π θ φ θ

×+ ⋅

[ ( , )– ( , )]

ei k k rp

ii i q

si iθ φ θ π φ

+ ( ) +( ) , ; ,R fq i qp i i i iθ θ π φ θ φ

×+ ⋅

[ ( , )– ( – , )]

ei k k rp

ii i q

si iθ φ π θ π φ

E r

er

fqs

ikr

qp i i i i( ) = +( )

π θ π φ θ φ– , ; ,


Figure 2. Magnified image of multi-layer coating, showingalternating layers of Mo and Si. The dark and light bands areMo and Si, respectively.

(a) (b)

Reflection mask Multilayer, coated condenser optics

Laser-producedplasma

High-powerlaser

Multilayer, coatedimaging optics

Figure 1. The EUV lithographic process. (a) EUV light is guided toward the mask using multi-layer-coated reflective optics. After the lightpasses through the mask and the multi-layer imaging optics, the circuit patterns on the mask are transferred onto Si wafers with reducedimage size. The ion beam deposition system (b) is used to produce these uniform low-defect-density multi-layer-coated reflective maskblanks. The system includes the substrate loader and the deposition chamber where the multi-layer coatings are applied.

430 Wang_qk 9/8/99 6:16 PM Page 4-24


where q and p are the polarization components (q, p= horizontal or vertical) of the scattered and incidentwaves, respectively; fqp is the scattering matrixelement; k i

p is the propagation vector of the incidentwave; k s

q is the propagation vector of the back-scattered wave; is the location of the scatterer;Rp(θi) and Rq(θi) are the Fresnel reflection coefficients.

The first-order solution in the above equationdescribes four major scattering mechanisms(Fig. 3). The first term in the equation is the directscattering from a particle (Fig. 3a). The secondterm is a single scattering from the scatterer,followed by a reflection off the boundary between airand the substrate (Fig. 3b). The third term is theopposite of the second term (Fig. 3c), and the fourthterm describes a reflection by the boundary followedby a single scattering from the particle and furtherfollowed by a reflection off the boundary (Fig. 3d).

r

×

+ ⋅ ,

[ ( – , )– ( , )]e E

i k k rpip

ii i q

si iπ θ φ θ π φ

+ ( ) +( ) ( ) , ; – ,R f Rq i qp i i i i p iθ θ π φ π θ φ θ

×+ ⋅

[ ( – , )– ( – , )]

ei k k rp

ii i q

si iπ θ φ π θ π φ The scattering matrix element fqp in the equa-

tion is calculated using the analytic method fordifferent defect shapes. For example, Rayleigh andMie scattering methods were applied to calculatethe scattering returns from spheres.4 Scatteringfrom cylindrical defects (Fig. 4a) was calculatedusing the finite cylinder approximation9 in whichthe induced current in the dielectric cylinder wasassumed to be the same as that of the infinitelylong cylinder of the same radius. This is a goodapproximation when the cylinder is long and thincompared to the wavelength. The scattered field isobtained by evaluating the field radiated from thisapproximate induced current source.

For disk-shaped defects (Fig. 4b), the returnswere calculated using the physical optics approxi-mation for elliptic disks,10 which assumes theinternal field inside the disc to be the same as thatof the infinitely extended dielectric layer of thesame thickness. Numerical full-wave techniques inboth the time and the frequency domain, includingANSOFTTM and TSARLITE,11 were used to validatescattering returns calculated with asymptotic/-analytical methods.

The scattering result from a cylindrical dielec-tric defect is shown in Figs. 5 to 7 as an example

FY 98 4-25

(a) (b)

(c) (d)

430 Fig 04 Art

Substrate Substrate

(a) (b)

Figure 3. Four majorscattering mechanismsconsidered part of thecharacterization ofEUVL defects: (a) directscattering from a scat-terer; (b) single scatter-ing from the scattererfollowed by reflectionoff the boundary; (c) surface reflectionfollowed by singlescattering; and (d) direct reflectionfrom the boundary,direct scattering fromthe scatterer,secondary reflection offthe boundary.

Figure 4. (a) Scattering from a cylindrical dielectric defect, calculated using the finite cylinder approximation; (b) scattering from a dielectricdisk-shaped defect, obtained using the physical optics approximation.

430 Wang_qk 9/8/99 6:16 PM Page 4-25


of the results we have obtained. Before the scatteringfrom a cylindrical defect on top of the substratewas calculated, analytic scattering results from avertical dielectric cylinder (without consideringthe ground effect) was first compared with the resultobtained from TSARLITE, a 2-D finite-differencetime-domain code.

The 2-D problem is equivalent to taking a cut-plane at some length along the cylinder. A TSARLITEsimulation of scattering from a cylinder is shownin Fig. 5.

In Fig. 6, scattering cross-sections of a verticaldielectric cylinder for HH and VV polarizations areplotted as a function of the scattering angle on thecut-plane. Scattering in the forward direction is at0°, while back-scattering is at 180°. The verticalaxis is the scattering cross-section of the cylinder inunits of µm2, plotted in dB.

The scattering cross-section, σqp, is defined interms of the incident field and the scattered electricfield described in the previous equation, as follows:

For the 2-D problem, instead of scattering area, thescattering width (obtained for the infinite structure)is defined:

The 3-D scattering cross-section of a truncated 2-Dstructure can then be calculated from the approxi-mate relationship12

σ π2

2

22Dqp

r

qs

pi

rE

E lim

| |

| |.=

→∞

σ πqpr

qs

pi

rE

E lim

| |

| |.=

→∞4 2

2

2

where is the length of the structure, λ is thewavelength, and τ is the tilt angle of the structuremeasured from broadside incidence.

The comparison between σHH, σVV calculatedfrom analytic approximation and σ3DHH, σ3DVVobtained from TSARLITE is shown in Fig. 6. Thelength and radius of the cylinder are 5λ and 0.1λ,respectively. The dielectric constant of the cylinderis 19 + i0.7. The comparison shows good agreementbetween the analytic and TSARLITE results.

Figure 7 shows the HH and VV back-scatteringradar cross-section of a cylindrical dielectric defectlying on top of the substrate as a function of theincident angle (measured from the vertical direc-tion). The scattering returns include the direct scat-tering from the cylinder, and the phase interactionsbetween the substrate and the cylinder. The lengthand radius of the cylinder are the same as in Fig. 6.The dielectric constant of the cylinder and substrateis 19 + i0.7.

This analytic solution is indicative of thesize/structure of the defect and will allow the experi-mental inspection to determine characteristics ofthe substrate anomalies.

Future Work

For the coming year, we will extend the models toinclude scattering from defects under multi-layercoatings. A combination of asymptotic/analytical

l

σσλ

ττ3

22

22Dqp

Dqp kk

sin( sin )

sin,=

l l

l


Figure 5. TSARLITE simulation of scattering from a dielectriccylinder, performed to validate the analytical approximation.

0 50 100 150 200 250 300 350-5

0

5

10

15

20

25

Scattering angle (degrees)

Scat

teri

ng

cro

ss-s

ecti

on

(d

B-

µ m

2 ) HH (analytic approx.)HH (TSARLITE) VV (analytic approx.)VV (TSARLITE)

Figure 6. Scattering from a vertical dielectric cylinder as afunction of the scattering angle on a cut-plane at some lengthalong the cylinder. The length and radius of the cylinder are5λ and 0.1λ, respectively. The dielectric constant of the cylin-der is 19 + i0.7.

430 Wang_qk 9/8/99 6:16 PM Page 4-26


methods and numerical techniques will be used tocalculate scattering from those defects. Numericaltechniques will be used to validate scatteringreturns calculated with asymptotic/analytical meth-ods. They will also be used in calculating scatteringfrom defects with more complex shapes, and inthose cases where the asymptotic/analytical meth-ods are no longer valid.

At the end of the year, the scattering model willbe extended to include scattering from defectsunder multi-layer coatings. The simulation resultsinclude scattering cross-section for detection ofdefects, as well as the scattered field. The lattercouples into lithography printing simulation codes.Simulation programs will be benchmarked withexperimental data.

Acknowledgment

The authors wish to thank M. Bujak for assis-tance with the TSARLITE results.

References

1. Vernon, S. P. , D. R. Kania, P. A. Kearney, R. A.Levesque, A. V. Hayes, B. Druz, E. Osten, R. Rajan,and H. Hedge (1996), “Reticle Blanks for ExtremeUltraviolet Lithography: Ion Beam Sputter Depositionof Low Defect Density Mo/Si Multilayers,” OSATopical Meeting on Extreme Ultraviolet Lithography,Vol. 4, G. Kubiak and D. Kania, eds, pp. 44-48.

2. Kearney, P. A. , C. E. Moore, S. I. Tan, S. P. Vernon,and R. A. Levesque (1997), “Mask Blanks for ExtremeUltraviolet Lithography: Ion Beam Sputter Depositionof Low Defect Density Mo/Si Multilayers,” Journal ofVacuum Science and Technology B, Vol. 15, (6), pp. 2452– 2454.

3. Lin, Y., and J. Bokor (1997), “ Minimum CriticalDefects in Extreme-Ultraviolet LithographyMasks,” Journal of Vacuum Science andTechnology B, Vol. 15, (6), pp. 2467– 2470.

4. Ishimaru, A. (1991), Electromagnetic WavePropagation, Radiation, and Scattering, Prentice Hall,Englewood Cliffs, New Jersey.

5. Kong, J. A. (1990), Electromagnetic Wave Theory,2nd ed., John Wiley and Sons, New York, New York.

6. Ruck, G. T., D. E. Barrick, W. D. Stuart, and C. K.Krichbaum (1970), Radar Cross Section Handbook,Vol. 1, 2, Plenum Press, New York, New York.

7. Tsang, L., J. A. Kong, and R. T. Shin (1985), Theory ofMicrowave Remote Sensing, John Wiley and Sons,New York, New York.

8. Bouche, D., F. Molinet, and R. Mittra (1997),Asymptotic Methods in Electromagnetics, Springer-Verlag, New York, New York.

9. Karam, M. A. , A. K. Fung, and Y. M. M. Antar (1988),“ Electromagnetic Wave Scattering from SomeVegetation Samples,” IEEE Transactions onGeoscience and Remote Sensing, November, Vol. 26,(6), pp. 799– 808.

10. Le Vine, D. M., A. Schneider, R. H. Lang, and H. G.Carter (1985), “ Scattering from Thin DielectricDisks,” IEEE Transactions on Antennas andPropagation, December, Vol. 33, (12), pp. 1410– 1413.

11. Kallman, J. S. , and R. J. Hawkins (1992), TSARLITEand BEEMER: Tools for Integrated Optic Design,Lawrence Livermore National Laboratory, Livermore,California (UCRL-JC-110403).

12. Skolnik, M. (1990), Radar Handbook, 2nd ed.,Chapter 11, McGraw-Hill, New York, New York.

FY 98 4-27

0 10 20 30 40 50 60 70 80 900

2

4

6

8

10

12

14

16

Incident angle (degrees)

Bac

ksca

tter

ing

cro

ss-s

ecti

on

(d

B-

µ m

2 )

HHVV

Figure 7. Back-scattering radar cross-section from a cylindricaldielectric defect lying on top of the substrate, as a function ofthe incident angle. The result takes into consideration thephase interactions between the cylinder and the substrate. Thelength and radius of the cylinder are 5λ and 0.1λ, respectively.The dielectric constant of the cylinder is 19 + i0.7 .

430 Wang_qk 9/8/99 6:17 PM Page 4-27

uclear and Electromagnetic Radiation SimulationTools for Dual-Revalidation of the Stockpile


Introduction

The goal of this project is to obtain, install, vali-date, use, and improve simulation software tools forpredicting system-generated electromagnetic pulse(SGEMP) effects in the stockpile dual-revalidationprogram. SGEMP is the transient electromag-netic pulse created by electrons emitted by asystem exposed to incident photons from largegamma fluences.

SGEMP can be external or internal. In the exter-nal case, the electromagnetic fields are produced onthe outer surfaces of the system. In the internalcase, the fields are produced within the system.SGEMP can cause upset, damage, and system fail-ure due to large currents burning out components.The presence of trapped or evolved gasses aroundthe system or inside cavities significantly modifiesSGEMP. Experiments indicate that breakdown of thespace charge layer occurs at low pressures, signifi-cantly increasing current flow. For this reasonvacuum simulations are not sufficient to evaluateworst case threats.

Approach

Our approach to this project has been to obtain,install, run, and test the codes MEEC, MCNP, andCEPXS. We have also obtained a dedicated Pentium-class PC with a removable hard drive and a state-of-the-art FORTRAN compiler and development studioto run these codes. We have contracted to have thepresently script-based MEEC code upgraded with a

GUI to make it much more user-friendly. In addition,we have compared results from the MEEC vacuumregime code with those from the TS3 electromag-netic particle-in-the-cell (EM-PIC) module of theMAFIA code.

MEEC is a set of 2- and 3-D EM-PIC codes forSGEMP in the vacuum, low-pressure, and atmos-pheric regimes. MCNP (Monte Carlo NeutronPhoton) is a 3-D, time-dependent code for electrongeneration due to photon and neutron incidence andtransport in materials. CEPXS solves the 1-D trans-port equation for coupled photon-electron cascadesfor energies greater than 1 keV. Since CEPXS is adeterministic, finite-difference code, it is lesscomputationally intensive than a Monte Carlo codeand will allow more efficient calculation of gener-ated electron fluences for input to the MEEC code.

Validation

As a validation of the MEEC vacuum regimecode, we have run a test problem consisting of anelectrically perfectly conducting right rectangularsolid within an electrically perfectly conductingright rectangular container. Two values of electroncharge emission from the top of the internal solid,6.33 × 10–8 C and 6.33 × 10–6 C have been simu-lated. We have also run the same problem with asimilar, but slightly different, computational gridwith MAFIA TS3.

The MAFIA geometry is shown in Fig. 1. Sincesymmetry is assumed about the z-axis, the MAFIAcalculation is confined to the first quadrant of the

FY 98 4-29

We have obtained and installed three codes for the assessment of system-generated electromag-netic pulse effects for dual-revalidation of the stockpile. We have run these codes and have begunvalidation of them. The good agreement between results from the MEEC vacuum regime code and theMAFIA code for the case of low-emitted charge gives us confidence that the MEEC provides accurateresults for non-space-charge-limited emission for simple internal geometries. Documents we haveobtained show good agreement between MEEC simulations and experiments for low- and high-pressure cases using simple geometries.

David J. Mayhall and Michael F. BlandDefense Sciences Engineering DivisionElectronics Engineering

435 Mayhall_qksae 7/27/99 7:39 AM Page 4-29


z-projected xy plane (positive x and positive y). Thexz and yz planes are symmetry planes. The domain ofthe inner solid is 0 ≤ x ≤ 0.5 m, 0 ≤ y ≤ 0.5 m, –0.5 m≤ z ≤ 0.5 m. The domain of the computational spacewithin the outer solid is 0 ≤ x ≤ 2 m, 0 ≤ y ≤ 2 m,–2 m ≤ z ≤ 2 m. Part of the spatially uniform MAFIAgrid is shown on the surfaces of the inner solid.

The time history of the emission is triangular,with a linear rise from 0 to 10 ns, followed by alinear fall from 10 ns to 20 ns to a null value. Theangular distribution of emitted particles about thez-axis in the plane parallel to the xy plane isuniform. The angular distribution of emitted parti-cles, as measured from 0 to 90° from the positivez-axis, is parabolic, with minima at 0 and 90° and amaximum at 45°. The spatial distributions of emittedparticles are uniform in the x and y directions. Thedistribution of emitted particles with kinetic energyis half-parabolic with the maximum at 1 keV, theminimum at 24 keV, and a maximum to minimum

ratio of 10. The spatial state of the electronmacroparticles emitted from the top surface of theinner solid at 21.6 ns, for 6.33 × 10–8 C of emittednegative charge, is shown in Fig. 1.

The MAFIA calculation uses one randomly emit-ting area. The MEEC calculation uses five emittingareas of extent, 0.1 m × 0.5 m, in the x and y direc-tions. Each calculation uses 4000 macroparticles.The MEEC grid is nonuniform, with finer spacingsnear the interfaces of the inner solid with thesurrounding vacuum space. In both calculations, theelectrons incident on the inner surfaces of the outercontainer and inner solid are totally absorbed.Because of restrictions on image plane locations,the MEEC calculation is performed in the third quad-rant of the z-projected xy plane (negative x andnegative y).

Figure 2 shows the spatial state of the particlesfor the MAFIA calculation at 21.6 ns for 6.33 × 10–6 Cof emitted charge. For this case, fewer particles are


Y

Z

X

Figure 1. MAFIA problem geometry and the electronmacroparticles at 21.6 ns for 6.33 x 10–8 C of emitted charge.

Y

Z

X

Figure 2. The MAFIA electron macroparticles at 21.6 ns for6.33 x 10–6 C of emitted charge.



situated within the bulk of the space between thetwo solids. More are confined near the surface of theinner solid. Space-charge-limited emission is thusevident in this second calculation. For this amountof emitted charge, many of the emitted particlesremain close to the inner solid, move down its sides,and then move across its bottom at z = –0.5 m.

Figure 3 shows a comparison of MEEC andMAFIA Ez-field waveforms at two nearby pointsabove the emitting top of the solid on the positivez-axis. The MEEC electric field point is at (0, 0,0.81 m); the MAFIA point is at (0, 0, 0.8125 m). TheMAFIA result is noisier and rises to a higher corre-sponding positive value than does the MEEC result.Some of the MAFIA noise may be due to the singleemitter used, as opposed to the five emitters used inthe MEEC calculation. The discrepancy in the valuesafter the negative peaks may be due to differences inthe two grids.

Figure 4 shows a comparison of MEEC andMAFIA Hx waveforms at two more nearby points.The MEEC field component is calculated at (0,–0.425 m, 0.525 m); the MAFIA component at (0,0.4375 m, 0.5625 m). These points lie on the x = 0symmetry plane, slightly above the emitting surfacenear the outer edges of the inner solid. Because theinitial axial current density in the space above theinner solid is in the negative z direction, the right-hand rule gives a negative Hx component for theMEEC simulation in the third quadrant and a positiveone for the MAFIA simulation in the first quadrant.

For ease of comparison, we have inverted theMAFIA result in Fig. 4. The MAFIA point is slightlycloser to a top edge of the inner solid (1.25 cm)and somewhat higher above the emitting surface(3.75 cm) in the positive z direction. The MAFIAwaveform is thus expected to be delayed withrespect to the MEEC waveform and to be some-what lower in amplitude. Figure 4 shows theMAFIA result to be delayed by about 1 ns duringthe fall toward the negative peaks. An apparentsmoothed value of the MAFIA negative peak is~–2.5 A/m, which is slightly lower in magnitudethan the MEEC negative peak of ~–2.55 A/m. TheMAFIA result is once again more noisy than theMEEC result.

Greater differences between the results of thetwo codes occur for 6 × 10–6 C of emitted charge.Although the shapes of the waveforms are generallysimilar, the MAFIA peaks are usually greater anddelayed. Some of the MAFIA Ez waveforms showstrong late time (>29 ns) discrepancies. The MAFIAwaveforms, especially for the magnetic field andcurrent density components, are also noisier. Thereasons for these discrepancies are not yet known.Some possibilities include computational differencesand object description differences. However,because a run with a more refined MEEC gridproduced results for the high-charge-emission casesimilar to those for the initial MEEC grid, we believethat MEEC results are good for space-charge-limitedemission in simple geometries.

FY 98 4-31

E z (

V/m

)

-1500

-100

-500

0

500

1000

1500

2000

MEECMAFIA

35x10-9302520Time (s)151050

Figure 3. Comparison of MEEC and MAFIA Ez components for6.33 x 10–8 C of emitted charge.

Figure 4. Comparison of MEEC and MAFIA Hx components for6.33 x 10–8 C of emitted charge.

Hx

(A/m

)

-3.0

-2.5

-2.0

-1.5

-1.0

-0.5

0

MEECMAFIA

35x10-9302520Time (s)151050



low- and atmospheric-pressure codes in simplegeometries. When available, we plan to test MEECwith the new GUI capability. To support the W76revalidation we plan to model the W76 weapon elec-tronic and radiation case geometry. Simulations ofcable-induced SGEMP currents will be calculatedand compared to test data.


Future Work

We plan to further investigate the accuracy of theMEEC vacuum and the MAFIA codes for cases ofspace-charge-limited emission. Benchmarkingagainst standard test problems or accepted analyticalsolutions is desired. We plan to validate the MEEC


elf-Effects in Expanding Electron Beam Plasmas


Introduction

When a pulsed electron beam hits a metal platewith sufficient energy, a volume of the metalbecomes ionized fluid that subsequently sprays outof the plate. Does this counterflow disrupt the tightfocus of the initial electron bunch, or later pulses ina train? This work aims to model the spatial distrib-ution of plasma speed, density, degree of ionization,and magnetization to address this question.

Progress

Numerical simulations of target flows are veryaccurate for the initial 2 µs. Beyond this point, therelative density is quite low, precision drops, andcalculation times are long. The initial solid-density,several-eV plasma expands to 1 cm and 10–4 relativedensity by 2 µs. Yet, a Faraday cup detector located25 cm from the target at LLNL’s experimental testaccelerator (ETA-II) facility, observes the flow afteran expansion of 50 µs. In addition, the Faraday cupsat ETA-II are immersed in the focusing magneticfield of the accelerator. There is a wide gap betweenwhat the Faraday cups see and what the simulationsprovide. An analytical model of the plasma flowhelps to connect the experimental observations withthe simulations of early times.

The expansion of the target plasma into thevacuum of the accelerator is so rapid that theionized portion of the flow departs from local ther-modynamic equilibrium downstream. In fact, wefound that when the temperature (in eV) in a parcelof fluid drops below VI × [(2γ – 2)/(9γ + 15)], whereVI is the ionization potential of the target metal (forexample, 7.8 eV for tantalum), and γ is the ratio of

specific heats, then the fractional ionization andelectron temperature in that parcel remain fixedduring its subsequent expansion. This effect is calledfreezing. For atoms, γ = 5/3 and the freezing temper-ature as defined here is VI/22.5.1,2

An electron beam penetrating the target and itsplasma will experience a radical change in thebalance it had in vacuum, between the pinchingforce of its own azimuthal magnetic field and thedisruptive electric field of its negative space charge.

The effects are as follows: 1) the target plasmacancels the electric field of the beam by an over-whelming charge density; 2) the plasma conductivitycreates an internal eddy current to counter themagnetization introduced by the transit of relativis-tic electrons; and 3) the plasma magnetizationchanges as electron beam heating alters the densitygradient and the magnitude of the conductivity.

This combination of effects could cause eitherpinching or expansion of a penetrating beam (asidefrom scattering) at different times during its transit.References 3 and 4 initiate the exploration of ther-mally-generated magnetization, while References 5,6 and 7 are guides to a deeper analysis.

Figures 1, 2, and 3 show a tantalum flow gener-ated by depositing an impulse of 5 J within a diame-ter of 0.8 mm and depth of 50 µm. The initialtemperature of this plasma was 9 eV.

Figures 1, 2, and 3 also show profiles of theheavy particle density, fluid and electron tempera-tures, and electron density, respectively, outside theplate at 300 ns. The leading edge of the flow hasreached 1.2 mm, the source volume within the platehas cooled to less than 1 eV, and the source plasmahas 2% of its original density.

FY 98 4-33

An analytical model of plasma flow from metal plates hit by intense, pulsed, electron beams aimsto bridge the gap between radiation-hydrodynamics simulations and experiments in the AdvancedHydrotest Facility at Lawrence Livermore National Laboratory (LLNL). This model also helps quantify theself-effect of the electron beam penetrating the target material, which is not addressed by simulations.

Manuel GarciaLaser Engineering DivisionElectronics Engineering

440 Garcia_qk 7/27/99 7:41 AM Page 4-33


The freezing effect is just visible in Fig. 2. Flowlater in time and downstream of this point will havea frozen ionization fraction (2 × 10–5) and electrontemperature (0.35 eV). An ideal probe at 25 cm fromthe plate would observe the arrival of plasma at51 µs, with an atom density of 2 × 1016 cm–3, anelectron density of 4 × 1011 cm–3, and a decrease toone quarter of these peak values by 140 µs.

Figure 4 shows profiles of the azimuthalmagnetic field (tesla) that arises within the tantalumcloud, just described, on being penetrated andheated by an electron beam pulse of 1000 A within a0.8-mm diameter for 30 ns. One profile in Fig. 4 isof local magnetic field values, the other is a runningaverage from the leading edge of the flow.

The magnetic field is generated thermally by thenonalignment of the gradients of electron tempera-ture and density.5,6 This thermally-generatedmagnetic field has the opposite polarity to the self-field of the electron beam (which is – 0.5 T in thisexample). Large local values arise where thedensity gradients are large: at the leading edge,and at the exit of the plate. In this example, theaverage thermally-generated field does not causethe electron beam to diverge because high localfields occur only in narrow zones.


6 x 1020

5 x 1020

4 x 1020

3 x 1020

2 x 1020

1 x 1020

Distance (mm)1.00.50

Part

icle

den

sity

(#/

cc)

0

0.4

0.3

0.2

0.1

Distance (mm)

Electrons

Fluid

1.00.50

Tem

per

atur

e (e

V)

0

3.5 x 1016

3 x 1016

2.5 x 1016

2 x 1016

1.5 x 1016

5 x 1015

1 x 1015

Distance (mm)1.00.50

Elec

tro

n d

ensi

ty (

#/cc

)

0

Distance (mm)

Local

Average

Mag

netic

fiel

d (T

)

10

4

3

1

2

0

10-5 10-4 0.001 0.01 0.1 1

Figure 1. Particle density at 300 ns. Figure 3. Electron density after 300 ns.

Figure 2. Electron and fluid temperatures after 300 ns. Figure 4. Local and average magnetic field along the flow.

440 Garcia_qk 7/27/99 7:41 AM Page 4-34


Future Work

Current work aims to improve the estimates ofthermally-generated magnetic fields by refining themany approximations required to describe such intri-cate physics in an analytical model. In particular, arealistic conductivity model σ (T, N)8 and a moreaccurate dynamic heating model for improved densitygradients are the focus of effort. Also, comparison tothe Faraday cup and interferometer data from ETA-IIcan help improve both model and experiment.

References

1. Garcia, M. (1997), “Splash flow from a metal platehit by an electron beam pulse,” LawrenceLivermore National Laboratory, Livermore,California (UCRL-ID-128660).

2. Garcia, M. (1997), “Frozen plasma within the flowfrom a metal plate hit by an electron beam pulse,”Lawrence Livermore National Laboratory, Livermore,California (UCRL-ID-126296).

3. Garcia, M. (1998), “On electromagnetic accelerationof material from a plate hit by a pulsed electronbeam,” Lawrence Livermore National Laboratory,Livermore, California (UCRL-JC-130448).

4. Garcia, M. (1998), “Electron beam expansion bytarget heating,” Lawrence Livermore NationalLaboratory, Livermore, California (UCRL-ID-131291).

5. Haase, R. (1969), Thermodynamics of IrreversibleProcesses, Addision-Wesley Publishing Company,London.

6. Haines, M. G. (1997), “Saturation mechanisms forthe generated magnetic field in a nonuniform laser-matter irradiation,” Phys. Rev. Lett. 78 (2), p. 254.

7. Craxton, R. S., and M. G. Haines (1978), “j × B accel-eration of fast ions in laser-target interactions,”Plasma Physics, Vol. 20, Pergamon Press Ltd.,Northern Ireland, pp. 487–502.

8. Redmer, R. (1998), “Electrical conductivity of densemetal plasmas,” submitted to Phys. Rev. E, 6.

FY 98 4-35

440 Garcia_qk 7/27/99 7:41 AM Page 4-35

ump-Induced Wavefront Distortion in Prototypical NIF and LMJ Amplifiers


Introduction

We are currently developing large-aperture ampli-fiers for the National Ignition Facility (NIF) andLaser Megajoules (LMJ) lasers at LawrenceLivermore National Laboratory (LLNL). These multi-segment amplifiers are the flashlamp-pumped,Nd:Glass type, and are designed to propagate anominally 36-cm-square beam. The apertures withina particular amplifier bundle are arranged in a four-high-by-two-wide configuration and use two sideflashlamp arrays and a central flashlamp array forpumping (Fig. 1).

As shown in Fig. 1, the slabs are oriented atBrewster’s angle and are pumped on both sides byarrays of flashlamps, denoted as central arrays(lamps that pump both slabs) and side arrays. Thegeometry of the amplifier results in one end of theslab being situated closer to the lamps than theother. Consequently, the amount of heat deposited inthe slab (primarily from the quantum defect of the

broad-band pump light) is uneven front-to-back aswell as side-to-side.

This uneven pumping results in a warping of thelaser slab, indicated by the dashed lines in Fig. 1.An initially plane wavefront incident on such a slabwill not remain planar upon exit. Ultimately, wave-front distortion is due to differences in the opticalpath (defined as the refractive index times the physi-cal path length) at one point in the aperture vs thatat another. We have seen one source of these opticalpath differences (OPDs), namely, the mechanicaldistortion of the laser slab. There is another sourceof OPD: the spatially-varying refractive index towhich temperature and stress contribute.

We have developed a model that takes all of theabove effects into account, and that allows us topredict the pump-induced wavefront distortion forthese large-aperture amplifiers. In this report, wedescribe various aspects of the model and presentcomparisons between the model and experimentaldata taken on AMPLAB, our amplifier test laboratory.

FY 98 4-37

In large-aperture laser amplifiers such as those envisioned for the National Ignition Facility (NIF)and Laser Megajoules (LMJ) lasers, the geometry is such that the front and back faces of the laserslab are heated unevenly by the pump process. This uneven heating results in a mechanical deforma-tion of the laser slab and consequent internal stresses. The deformation and stresses, along with atemperature-dependent refractive index variation, result in phase variations across the laser beam,so-called pump-induced wavefront distortions. These phase variations lead to beam steering whichmay affect frequency conversion as well as energy-on-target. We have developed a model that allowsus to estimate the pump-induced wavefront distortion for a given amplifier configuration as well asthe spatially-resolved depolarization. The model is compared with experiments taken in our amplifierdevelopment laboratory (AMPLAB).

Mark D. RotterLaser Engineering DivisionElectronics Engineering

Kenneth S. Jancaitis, Christopher D. Marshall, Luis E. Zapata, and Alvin C. ErlandsonLaser Science and TechnologyLaser Programs

Geoffroy LeTouze′ and Stephane SeznecCommissariat à l′Energie AtomiqueVilleneuve, St. Georges, France

445 Rotter_qk 7/27/99 7:43 AM Page 4-37


Progress

Pump-Induced Wavefront-Distortion Model

Description of Model. As mentioned in theintroduction, the non-uniform deposition of heat inthe laser slab is responsible for slab distortions andaberrations of any plane wavefront incident on it.Thus, the fundamental ideas may be summarized asfollows: the non-uniform heat deposition results in adistortion of the laser slab and accompanyingstresses. The distortion of the laser slab, in conjunc-tion with the temperature and stress-induced refrac-tive index variations result in the OPDs and conse-quent wavefront distortion.

To calculate the various effects listed above, weuse a suite of computer codes: TOPAZ3D1 to calcu-late the temperature distribution within the laserslab, NIKE3D2 to calculate the displacements andstresses from the temperature field given byTOPAZ3D, and OPL, which calculates the OPDs,given the results from NIKE3D. The codes NIKE3Dand TOPAZ3D are 3-D finite element analysis codeswhich have been in use at LLNL for more than tenyears. The optics code OPL is an in-house codebased on the BREW code.3

Determination of Temperature. The first stepin calculating the wavefront distortion is to deter-mine the temperature distribution within the laserslab. To do this, we need to specify the thermalsource function as a function of position and time. Ingeneral, this thermal source function is an arbitraryfunction of position and time. For purposes of thismodel, however, we have assumed a separablesource function, that is:

Q x y z t As x g y f z, , , ( ) = ( ) ( ) ( )[ 0

, (1)

where s(x) denotes the vertical variation of thepump profile; g0,h(y) denotes the horizontal variationof the pump profile at z = 0, h; f(z) denotes thepump profile through the thickness of the slab; andQec is the thermal source term for the edgecladding. The units for this source function areW/cm.3 In Eq. 1, h is the thickness of the slab and Ais a constant multiplier. Each of these terms willnow be described.

In the multi-segment amplifiers envisioned foruse in the NIF, there is strong vertical symmetry. Theflashlamps are oriented vertically, and there aresilver-coated metal reflectors at the top and bottomof the pump cavity. As a result, we have taken thefunction s(x) to be a constant. In reality, the reflec-tors are not perfect, and so there is a slight roll-offin pump light at the extreme top and bottom of thepump cavity. We have found that this is a smalleffect insofar as calculating the pump-induced wave-front distortion is concerned, and so have elected tokeep s(x) a constant.

As mentioned above, the geometry of the pumpcavity results in a roll-off of the pump radiationfrom one side of the aperture to the other. We usedour 2-D+ ray-trace code4 to calculate the distribu-tion of pump light across the laser slab. An exam-ple of the functions so obtained is shown in Fig. 2.Because these functions can be rather convoluted,no curve fitting is done. Instead, TOPAZ3D usesg(y) as is, and interpolates to return values for Qfor any y.

The thermal energy deposition through the thick-ness of the slab is given by the function f(z). Thisfunction is calculated using our Lamp Model code,5

which calculates the energy deposition profile as a

+ ( ) ( )] ( ) + – g y f h z u t Qh ec


Figure 1. Plan view ofmulti-segment ampli-fier showing geome-try of Brewster-anglelaser slabs. Surfacedistortions (greatlyexaggerated) arecaused by unevenpumping.

Brewster-angle slab

Flashlamps

Surface distortions

445 Rotter_qk 7/27/99 7:43 AM Page 4-38


function of slab geometry, doping density, and lampoperating level. The profile so obtained is a spectrally-integrated thermal energy deposition profile, whichwe then fit to a double exponential:

, (2)

where µ1, µ2, and c are the fit coefficients. A plot ofthe energy deposition profile and the correspondingfit is shown in Fig. 3.

To describe the temporal behavior of the pumppulse, we use an analytic expression for u(t). Wefirst need to describe the temporal behavior of theelectrical input power to the lamp. Since the flash-lamp is a nonlinear circuit element, the acutal pulseshape is described by a nonlinear differential equa-tion.6 We have found, however, that an excellentapproximation to the shape is given by the function:

, (3)

where a and τ are fit parameters. A plot of the elec-trical input power as determined from a numericalintegration of the circuit equation and the approxi-mation given by Eq. 3 is shown in Fig. 4. With theelectrical input power so determined, we then calcu-late the optical output power from the flashlamp.The output power, u(t), may be calculated from thefollowing equation:7

, (4)

where η(u) is the instantaneous radiant efficiency ofthe flashlamp (corrected for arc-expansion effects),and τR is radiative recombination time of the plasma(~30 µs). Since η(u) is a nonlinear function of u,Eq. 4 is likewise nonlinear.7 We have found,however, that an excellent approximation is to takeη(u) to be a constant, parametric in the pulsewidth:

, (5)

where τ10 is the full-width tenth-maximum time of theelectrical input power pulse. In Fig. 5 we show thecomparison between the numerical solution to Eq. 4 andthe solution using the approximation given by Eq. 5. Wesee that over the time range specified, the agreement isquite good. Using Eqs. 3 to 5, an analytic expression foru(t) may be obtained.

At this point, we do not have a good ab initiocalculation for the heat deposited in the laser slab.Present calculations disagree with measurements by

200 50010 < <τ µs

η τu( ) = + × . . –653 2 33 10 4

10

du dt u p t u t R/ – = ( ) ( ) ( )[ ]η τ

p t t t a( ) exp – – ∝ ( )

2 2τ

f z e c ez z( ) – –∝ +µ µ1 2

FY 984-39

0-0.4 -0.3 -0.2 -0.1 0 0.1 0.2 0.3 0.4

0.2

0.4

0.6

0.8

1.0

1.2

Pum

p ir

rad

ian

ce (

a. u

.)

Distance along slab (m)

Back surface

Front surface

Figure 2. Pump irradiance profiles—AMPLAB diamond configuration.

0

6

8

10

0 0.5 1 1.5 2.52 3 3.5 4

2

4

Dep

osi

ted

en

erg

y (a

. u.

)

Depth (cm)

4.2% doping

f(z) ∝ Exp[-µ1z] + a Exp[-µ2z]

Fitting function, f(z)

Calculated from lamp model code

SlabFit

Figure 3. Energy deposition profile through slab and fit.

0

0.2

0.4

0.6

0.8

1.0

1.2

0 0.1 0.2 0.3 0.4 0.5 0.6

Lam

p in

put

po

wer

(a.

u.)

Time (ms)

fx = 0.2

p(t) ∝ t Exp[-(t-a)2/τp

2]

Fit

Measured

FitMeasured

Figure 4. Measured electrical input power to lamp and fit.

445 Rotter_qk 7/27/99 7:43 AM Page 4-39


about a factor of two. The cause of this discrepancyis currently not known. Thus, we have included ascale factor, A in Eq. 1, to scale our results toexperimental measurements. It should be pointedout that once this factor is determined (for example,from AMPLAB measurements), then that factor isheld constant for all succeeding calculations.

The last term in Eq. 1 represents the heatdeposited into the edge claddings that surroundthe laser glass. This term, Qec, is composed oftwo parts:

, (6)

where Qec,pump and Qec,ASE represent the heatdeposited into the edge claddings by the pump andby amplified spontaneous emission (ASE), respec-tively. The source term is broken up in such amanner because the time dependence of the twoparts is different. For Qec,pump the time dependenceis just u(t), described above. For Qec,ASE the timedependence is different due to the fact that the peakof the ASE pulse occurs at the time of peak gain, notat the time of peak pump power. The relationshipbetween the output power from the flashlamp andthe stored energy density, ρ(t), is given inReference 7. Once ρ(t) is determined, it may beshown that the time dependence of the ASE, φ(t),may be written as:8

, (7)φ ρ ρt t a b t( ) ∝ ( ) + ( )[ ]

exp – 1

Q Q Qec ec pump ec ASE , ,= +

where a and b are constants. With Eq. 7 describingthe time dependence of the ASE, we then have, forexample, for the edge cladding at x = constant:

, (8)

where β is the edge-cladding absorption coefficient,and Γ is the incident fluence. At present, we do nothave an accurate ab initio calculation of the ASEfluence on the edge cladding. However, based onmeasurements with the Beamlet laser, we estimatea fluence of 4 J/cm2 for ASE and another 2.5 J/cm2

due to the pump light.Equations 1 to 8 are used in TOPAZ3D to deter-

mine the temperature distribution in the laser slab.Due to the shortness of the pump pulse (a fewhundred µs), adiabatic boundary conditions are usedon all faces of the slab. The result of this calculationis used in the code NIKE3D to calculate the displace-ments and stresses, as described in the next section.

Determination of Displacements and Stresses.As mentioned above, we use the temperature distri-bution in the slab to calculate the displacements andstresses. Since the displacements are very small (onthe order of 1 µm), we are in the linear elasticregime and the problem is a standard one in thermo-elasticity. Consequently, we use a thermo-elasticmaterial model and specify Poisson’s Ratio, Young’sModulus and the thermal expansion coefficient.9

These parameters are taken to be constants, inde-pendent of temperature, insofar as the maximumtemperature rise is on the order of 1 °C.

Q t eec ASE

x,

– ∝ ( )φ β βΓ


0

1

2

3

4

5

6

0

1

2

4

5

6

7

8

3

0

1

2

4

5

6

7

3

0

1

2

3

4

5

Ou

tpu

t p

ow

er (

a. u

.)

0 0.1 0.2 0.3 0.4 0.5Time (ms)Time (ms)

0 0.1 0.2 0.4 0.6 0.7

Pout-exactPout-approx

Radiant efficiency = 70% Radiant efficiency = 77%

Pout-exactPout-approx

Ou

tpu

t en

erg

y (a

. u

.)

Ou

tpu

t p

ow

er (

a. u

.)

Ou

tpu

t en

erg

y (a

. u

.)

0.5 0.80.3

Figure 5. Comparison of exact and approximate solutions to flashlamp output equation.

445 Rotter_qk 7/27/99 7:43 AM Page 4-40


At this point in the calculation, the mechanicalboundary conditions for the slab are specified. Inreality, the slab sits on a Marcel spring—a sinu-soidally varying metal strip. While the capabilityexists to model the spring as it actually exists, wehave found that a satisfactory substitution is to havethe slab sit on a region of metal one element thick.The nodes in the glass material are joined to thenodes in the metal so no slipping can occur. Thebottom of the metal region is simply supported, andwe also fix the displacements at two additionalcorners to eliminate rigid-body motion. The rest ofthe slab surfaces are assumed to be free.

NIKE3D calculates the displacements andstresses as a function of time during the course ofthe pump pulse. Typically, the code is run up to thetime of peak gain as the wavefront distortion at thattime is what is usually requested. However, it is asimple matter to run the code for times longer thanthe time of peak gain to compare with experiments.

Determination of OPDs. The last part of thecalculation involves computing the OPDs throughthe laser slab. For this, we use our in-house codeOPL. The optical path length of a ray through theslab may be written as:

, (9)

where n is the (spatially-varying) refractive indexand s is the distance along the ray path. There aretwo main sources of OPD in the laser slab: 1) varia-tions in path length caused by mechanical motion ofthe slab, and 2) variations in path length caused byrefractive index changes.

The variations in path length caused by mechani-cal motion of the slab are due to the spatially-varyingdisplacements calculated in NIKE3D; that is, a pointx, y, z on the slab is translated to:

. (10)

We take two effects into account to calculate thespatially-varying refractive index: 1) the variation ofrefractive index with temperature, and 2) the varia-tion of refractive index with stress (stress-opticeffect), that is:

, (11) + ( ) ( ) / , ,dn d x y zσ σ∆

n x y z n dn dT T x y z, , / , ,( ) = + ( ) ( )0 ∆

z z w x y z t , , ,→ + ( ) y y v x y z t , , ,→ + ( ) x x u x y z t , , ,→ + ( )

OPL , ,= ( )∫ n x y z ds

where n0 is the isotropic refractive index, dn/dTdenotes the change of refractive index with tempera-ture, and we have symbolically written the change inrefractive index due to stress as dn/dσ.

Note that in general, ∆T and ∆σ are functions oftime. However, for the purposes of calculating theOPD, we select one point in time for the calculation.This is permissible since the time duration of thelaser pulse is at most 20 ns. On this time scale, allthermal and mechanical motion is frozen.

The sequence of events in calculating the OPD isas follows. The OPL code reads in the finite-elementgeometry from the NIKE3D plot file. We then gothrough the mesh and break up each finite-element“brick” into six four-node tetrahedra and generate aconnectivity matrix for these tetrahedra. We thenuse Eq. 11 to calculate the refractive index at eachnode in the mesh. Within each tetrahedra, welinearize the refractive index:

. (12)

The four unknowns in Eq. 12 are uniquely deter-mined by the values of the refractive index at thefour nodes of a given tetrahedron. With the refrac-tive index linearized as in Eq. 12, we can thenanalytically solve the Eikonal equation10 for the raypath within a tetrahedron:

, (13)

where s is the distance along the ray path and r isthe position vector of the ray. The connectivitymatrix helps us determine which tetrahedron the raywill enter, and consequently at which nodes to eval-uate the refractive index to calculate the unknownsin Eq. 12. We then track the ray as it propagatesthrough all the tetrahedra, all the while accumulat-ing the distance that the ray propagates.

In addition to calculating the OPD for the ray, wecan also calculate the depolarization a ray experi-ences as it propagates through the optic. We do thisby assuming each tetrahedron acts as a linearretarder, and calculate the Jones matrix11 for eachtetrahedron. The final amount of retardation (andhence depolarization) is given by the product of allthe individual Jones matrices for a given ray path.

A typical result from this calculation is shown inFig. 6. The calculation was performed for theAMPLAB amplifer in the diamond configuration. InFig. 6a we show the OPD for all effects (displace-ment, temperature, and stress) combined. It is often

dds

ndds

nr

= ∇

n x y z a bx cy dz( , , ) = + + +

FY 98 4-41

445 Rotter_qk 7/27/99 7:43 AM Page 4-41



0.20

0.15

10

0–10

–20 –100 10

2030

–30

0.05

0.00

OPD

(w

aves

/sla

b/p

ass

at 1

µ)

–20

0.10

Horizontal position (cm)

(a)0.20

0.15

10

0–10

–20 –100 10

2030

–30

0.05

0.00

OPD

(w

aves

/sla

b/p

ass

at 1

µ)

–20

0.10


(b)

Vertical position (cm)


0.20

0.15

10

0–10

–20 –100 10

2030

–30

0.05

0.00

OPD

(w

aves

/sla

b/p

ass

at 1

µ)

–20

0.10


(c)0.20

0.15

10

0–10

–20 –100 10

2030

–30

0.05

0.00

OPD

(w

aves

/sla

b/p

ass

at 1

µ)

–20

0.10


(d)



0.15

–20 –100 10 20

0.05

0.00

Dep

ola

riza

tio

n (

%)


–30

0.10

100

–10–20

–30


30

(e)

Figure 6. OPD calculated for the AMPLAB diamond configuration, fx = 0.2: (a) all effects; (b) displacement effects only; (c) dn/dTeffects only; (d) stress effects only; and (e) depolarization.

445 Rotter_qk 7/27/99 7:44 AM Page 4-42


useful to examine each component indivudually, andthis is done in Figs. 6b, c, and d. In Fig. 6b weshow the contribution to the OPD due solely tomechanical deformation.

Comparison with Fig. 6a shows that forAMPLAB, the wavefront distortion is due mainlyto the mechanical deformation of the slab. InFigs. 6c and d we show the contriubtion to theOPD from temperature and stress effects on therefractive index. As may be seen, these effectsplay a relatively minor role in determining the overall OPD. Finally, in Fig. 6e we show theP-to-S depolarization.

As expected, the greatest amount of depolar-ization occurs in the corners, where the twopieces of edge cladding meet. It is there wherethe greatest amount of stress occurs.Nevertheless, the overall amount of depolariza-tion is small, well within its specification of0.05% averaged over the aperture.

Error Analysis. In this section, we estimatethe error in our calculation of the OPD. For theamplifier conditions considered in this report,the dominant contribution to the OPD is themechanical deformation of the laser slab (seeFigs. 6a and b). Consequently, it makes sense toclosely analyze the uncertainties associated withmechanical motion.

For simplicity, assume the laser slab is asimply-supported thin plate, with the thin dimen-sion along the z-axis. We will neglect any timedependence in this analysis. It may be shownthat the equation for w, the displacement in thez-direction, is given by:12

, (14)

where the thermal moment, defined as:12

∇ ( ) =( )

2

1w x y

M x y

vT

, –,

–

FY 98 4-43

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

0.4

-20 -15 -10 -5 0 5 10 15 20

MeasuredCalculated

Phas

e (w

aves

/sla

b/p

ass

at 1

µ)


t = tPeak gain

Calculated

t = tPeak gain

Measured

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

0.4

-20 -15 -10 -5 0 5 10 15 20Horizontal position (cm)

(a) (b)

0.4

-20 -15 -10 -5 0 5 10 15 20

Measured

Phas

e (w

aves

/sla

b/p

ass

at 1

µ)


t = tPeak gain

t = tPeak gain

Interpolated

-0.4

-0.3

-0.2

-0.1

0

0.1

0.2

0.3

0.4

-20 -15 -10 -5 0 5 10 15 20Horizontal position (cm)

-0.6

-0.4

-0.2

0

0.2

(c) (d)

Calculated Calculated

Figure 7. Measuredand calculated wave-front, horizontalcomponent—AMPLAB: (a)diamond configura-tion; (b) X configura-tion. Error bar showstypical error formeasurements. (c)Three-slab-longconfiguration; (d)interior configura-tion—interpolatedfrom 3-slab-long,diamond, and Xresults.

445 Rotter_qk 7/27/99 7:44 AM Page 4-43

(15)

is the source function for the displacement. InEqs. 14 and 15, ν is Poisson’s Ratio, α is the ther-mal expansion coefficient, and E is Young’s modulus.

On the time scales of interest, one may neglectdiffusion. Consequently, T(x, y, z) ∝ Q(x, y, z). Ifone substitutes Eq. 1 into Eq. 15 using Eq. 2, onefinds that the thermal moment is proportional tothe difference in pump profiles, that is, MT ∝ g0(y)– gh(y). Consequently, small uncertainties in thevalues of g0 and gh can lead to large uncertaintiesin the thermal moment, and hence the amount ofdeformation. Since the steering of the laser beamis driven by the curvature of the laser slab, itfollows that the phase front, which is the integralof the beam-steering, is proportional to the gradi-ent of the displacement, or the integral of thethermal moment.

With our 2-D ray-trace code, we can match thegain profile across the aperture to within 1%.Because of ASE within the laser slab, we can varythe pump profile by 2% and still be within 1% in thegain coefficient. Consequently, if we take the pumpprofiles shown in Fig. 1, and assume a worst-caseuncertainty of 2%, it can be shown that the variationin the peak-to-valley value for the phase front can beas much as ±15%.

Comparison with AMPLAB Experiments

In this section, we present comparisons ofexperiments performed in AMPLAB with themodel described in the previous section. Unlessotherwise mentioned, all comparisons are doneat the time of peak gain at an explosion fractionof 0.2. As shown in Fig. 6, the OPD is calculatedover the entire aperture. However to facilitatemaking comparison with the data, we shall showhorizontal lineouts of calculations and experi-ments. These lineouts were taken at the verticalmidplane of the aperture.

In Figs. 7a and b, we show the comparisonbetween the calculated and measured phase front(essentially the negative of the OPD) for thediamond and X configurations in AMPLAB. Also indi-cated on the experimental curve is a typical value

M x y E z h T x y z dzT

h, – / , ,( ) = ( ) ( )∫α 2

0

for the error in the measurement. As may be seen,there is excellent agreement in both configurationsover the entire aperture.

In Figs. 7c and d, we show the calculated andmeasured phase front for the 3-slab-long andinterior configurations, respectively. As indicatedon the graph, the data for the interior configura-tion was interpolated from the measureddiamond, X, and 3-slab-long data using thefollowing algorithm:

, (16)

where φ3 is the measured 3-slab-long phase inwaves and all other phases are in waves/slab/pass.As can be seen from the figure, there is excellentagreement between calculation and measurementover the entire aperture.

Another check on the model is to calculate thewavefront distortion at times other than the timeof peak gain. The results of these calculations,and comparison with the measurement, areshown in Figs. 8a to f (Fig. 8b is repeated forease of comparison).

In these experiments, we measured the promptwavefront distortion at 100, 200, 300, and 500 µsafter the time of peak gain. As may be seen, theamount of wavefront distortion continues to increaseafter the time of peak gain up to 500 µs. In fact, thepeak-to-valley value of the wavefront distortion isabout three times greater at tpeak gain + 500 µs thanat tpeak gain. The agreement of the model with themeasurement is excellent, matching both the magni-tude and the shape of the wavefront at all times.

Summary and Conclusions

We have presented the results of detailed analy-sis and modeling of the AMPLAB data, summarizedin Table 1.

We have also presented a description of ourprompt pump-induced wavefront model. This modelcalculates the wavefront distortion due to mechani-cal deformation, and refractive index changes due totemperature and stress. We have benchmarked thecode against AMPLAB measurements and will beusing it to predict the wavefront distortion for theNIF amplifiers.

φ φ φ φi d x – – = 3 2



445 Rotter_qk 7/27/99 7:44 AM Page 4-44


FY 98 4-45

Phas

e (w

aves

at

1µ)

-20 -15 -10 -5 5 10 15 20-0Horizontal position (cm)


(a) (b)

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

MeasuredCalculated

Phas

e (w

aves

at

1µ)



(e) (f)

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

Phas

e (w

aves

at

1µ)



(c) (d)

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

MeasuredCalculated

MeasuredCalculated

MeasuredCalculated

MeasuredCalculated

MeasuredCalculated

Figure 8. Measuredand calculated wave-front, horizontalcomponent—AMPLAB, diamondconfiguration:measurements takenat (a) t = tpeak gain;(b) t = tpeak gain + 0.1ms for fx = 0.2; (c) t= tpeak gain + 0.1 ms;(d) t = tpeak gain + 0.2ms; (e) t = tpeak gain +0.3 ms; and (d) t =tpeak gain + 0.5 ms.

445 Rotter_qk 7/27/99 7:44 AM Page 4-45


References

1. Shapiro, A. (1985), “TOPAZ3D—A 3-D FiniteElement Heat Transfer Code,” Lawrence LivermoreNational Laboratory, Livermore, California (UCID-20484), August.

2. Maker, B. N. (1995), “NIKE3D—A Nonlinear, Implicit3-D Finite Element Code for Solid and StructuralMechanics,” Lawrence Livermore NationalLaboratory, Livermore, California (UCRL-MA-105268,Rev. 1), April.

3. Doss, S., and R. Gelinas (1986), “Mathematics andPhysics of the BREW Code,” Lawrence LivermoreNational Laboratory, Livermore, California (UCRL-50021-86), pp. 7-132.

4. LeTouze′, G., O. Cabourdin, J. F. Mengue, M. Rotter,and K. Jancaitis (1996), “Shaped Reflectors for PumpCavities,” 2nd Annual Conf. Solid State Lasers forApplication to ICF, Limeil, France.

5. Jancaitis, K. (1986), “Flashlamp Modeling,”LawrenceLivermore National Laboratory, Livermore, California(UCRL 50021-86), p. 6-3.


6. Markiewicz, J. P., and J. L. Emmett (1966),“Design of Flashlamp Driving Circuits,” IEEEJ.Q.E., QE-2, p. 707.

7. Powell, H. T., A. C. Erlandson, K. S. Jancaitis, and J.E. Murray (1990), “Flashlamp Pumping of Nd:GlassDisk Amplifiers,” SPIE 1277, p. 103.

8. Jancaitis, K. S. (1993), “Disk AmplifierPerformance Model,” Lawrence Livermore NationalLaboratory, Livermore, California (UCRL-LR-105820-88/89), p. 6-12.

9. Stokowski, S. E., R. A. Saroyan, and M. J. Weber(1981), “Laser Glass - Nd-doped GlassSpectroscopic and Physical Properties, V1,” M-095, Rev. 2, V. 1, Lawrence Livermore NationalLaboratory, Livermore, California.

10. Born, M., and E. Wolf (1980), “Principles of Optics,6th ed.,” p. 122, Pergamon Press, New York, New York.

11. Azzam, R. M. A., and N. M. Bashara (1987),“Ellipsometry and Polarized Light,” North-Holland,Amsterdam, p. 488.

12. Boley, B., and J. Weiner (1960), “Theory ofThermal Stresses,” Ch. 12, John Wiley and Sons,New York, New York.

Table 1. AMPLAB data analysis.

Measured φhoriz. Calculated φhoriz.Configuration (waves/slab/pass) (wave/slab/pass - LG-770)

Diamond 0.22 ± .03 0.21 ± .03X 0.29 ± .03 0.27 ± .04Interior 0.18 ± .03 0.16 ± .02

445 Rotter_qk 7/27/99 7:44 AM Page 4-46

arallel Algorithm Development for Computational Mechanics


Introduction

This report features five parallel algorithm devel-opment projects that are part of our plan for imple-menting a full structural mechanics capability onour Accelerated Strategic Computing Initiative(ASCI) parallel computers.

Progress

Parallel Contact Algorithm with Load Balancing

Partitioning methods are the key to developingefficient algorithms for parallel computers.Generally, in designing parallel algorithms for a finiteelement method, the mesh is divided up by assigningelements to processors so that the calculation of theinternal forces is balanced across the processors.However, some applications in nonlinear mechanics

involve calculating contact at material interfaces thatmay represent as much as fifty to eighty percent ofthe total computation time.

For example, timing results for a benchmarkcrashworthiness simulation show that nearly eightypercent of the total time is used for the contactcalculation. In the algorithm described below twopartitions are used. The first is a partitioning of themesh and the second is a method for localizing andpartitioning the contact surfaces.

In our previous work1 we described a paralleltechnique for treating contact interfaces when thesurface motion remained in a localized region ofthe mesh throughout the dynamics of the calcula-tion. In the more general case it is not possible todefine a priori which surfaces will come intocontact. This more general form of contact,referred to here as arbitrary contact, will occur ifmaterial surfaces undergo folding or large defor-mation, or if objects on the grid are in motion.

FY 98 4-47

We have successfully implemented dynamic load balancing algorithms in the automatic contactand material erosion algorithms in the ParaDyn program. These algorithms provide an efficientmethod for treating arbitrary interface motion and extend the applicability and scalability of ParaDynsimulations to large problems of interest. ParaDyn algorithms for partitioning, parallel contact, andload balancing can be applied to parallel implicit methods. The most challenging algorithm researchforeseen for future parallel models is the development of an efficient linear solver for both implicitsolid mechanics and fluid models. New areas of development in particle continuum methods andparallel visualization are also described.

Carol G. Hoover and Robert M. FerenczDefense Technologies Engineering DivisionMechanical Engineering

Anthony J. De Groot and Robert J. SherwoodDefense Sciences Engineering DivisionElectronics Engineering

Edward ZywiczNew Technologies Engineering DivisionMechanical Engineering

Yuen L. Lee and Douglas E. SpeckComputer Applications DivisionComputations

450 Hoover_qk 7/27/99 7:46 AM Page 4-47


Hence, a localization technique must be used. Thearbitrary contact algorithm currently implementedin DYNA3D2,3 uses a localization technique that isconveniently extended into a parallel implementa-tion. This algorithm is a node-on-patch method. Thesteps are as follows:

1. For each surface node, find a second surfacenode which is the nearest node in the remain-ing set of surface nodes;

2. Construct the facets connected to the secondsurface node; and

3. Determine whether the first surface node pene-trates a facet connected to the second node. Ifso apply a contact force to prevent penetration.

The first step uses a grid of cubes, also referredto as buckets, to localize the search for nearestsurface nodes. Surface nodes are sorted into thebuckets and the search for the closest node iscarried out over 27 buckets in three dimensions.The side length of the buckets is set to the maxi-mum diagonal length of a face, plus an incrementallength based on the velocity of the nodes formingthe diagonal (Fig. 1a). This characteristic length isused so that at least one of the surface nodes asso-ciated with a penetrated patch is always containedin one of the 27 buckets in the set over which thesearch is conducted.

The parallel implementation of the arbitrarycontact algorithm partitions the contact surfaces byallocating the buckets to processors. The bucketsare added to the processors so as to approximatelyequalize the total number of surface nodes in eachprocessor. One overlap bucket is needed acrossprocessors so that surface nodes in the centerbucket of a strip of three buckets will find itsconnected patches in either the adjacent left or rightbucket (Fig. 1b).

The allocation of buckets to processors is carriedout by using 1-, 2-, or 3-dimensional geometriclayout of processors (Fig. 1c). This corresponds toallocating buckets which are slices of the grid in onedimension, rectangular pipes in two dimensions, andcubes in three dimensions. The algorithm is currentlyimplemented based on a fixed maximum number ofbuckets, Nmax, for the full problem. The processorgeometry is selected based on whether the number ofbuckets for a particular processor geometry is lessthan Nmax. If more than one processor geometrysatisfies this condition, then the processor geometrywith the highest dimensionality is selected.

The use of two partitions (one for the full meshand a second for contact surfaces) requires commu-nication of the contact force values from the contactpartition to the mesh partition. After this communi-cation step, the total nodal forces are computed inprocessors associated with the mesh partition.Similarly, the time integration to compute the veloci-ties and coordinates are calculated for nodal pointsin the mesh partition. After the time integrationstep, the updated nodal coordinates and velocitiesare communicated back to the processors contain-ing the nodes in the contact partition.

The load balancing of the contact calculationsoccurs at time intervals corresponding to the timeinterval over which rebucketing occurs in the case ofa single processor computer. This interval corre-sponds to the time required to move a surface nodeacross the length of a bucket. The load balancing iscarried out in parallel using parallel reduction oper-ations which provide each processor with theglobal minimum and maximum for the geometry,the bucket structure, and the number of nodes ineach bucket.

All processors then compute the rebucketing onthe full grid. This is followed by a communicationstep in which each processor sends and receivessurface nodes and patches corresponding to thebuckets in the new contact partition.

Reliably detecting contact is a difficult task.Figure 2 illustrates the deformation of a sheet ofshell elements enclosed by moving stonewalls form-ing the faces of a cube. Contact detection for thisproblem uses the new method for treating contactwhich accounts for the thickness of shell elements.This problem runs correctly using 64 processors.

The material erosion algorithm in DYNA3D usesan algorithm formulation similar to the arbitrarycontact described above and consequently wasmerged into the arbitrary contact algorithm. Theerosion algorithm has been implemented in parallelduring this last year. Figure 3 illustrates the erosionalgorithm for a small benchmark similar to the


c )

a ) b )

VVd

1 2 3 4 1 2 3 43 2

Figure 1. Illustration of the three steps in the load balancing ofarbitrary contact surfaces. a) The bucket length is set to themaximum diagonal length of a face, plus a predicted incre-mental length based on the velocity along the diagonal; b) aset of buckets, in this case 4 1-D strips, is divided among twoprocessors with one overlapping strip at the processor bound-ary; c) the grid of buckets is divided among processors bydividing the grid into 1-D slices, 2-D pipes, or 3-D cubes. Theexamples above show 4-, 8-, and 16-processor allocations ofthe same grid using slices, pipes, and cubes, respectively.

450 Hoover_qk 7/27/99 7:46 AM Page 4-48


model in large production calculations of a hyper-velocity impact on a reentry vehicle.

Linear Solvers for Implicit Methods

Essentially all the parallel algorithmic strategiesimplemented within ParaDyn will be directly applic-able to our implicit mechanics code, NIKE3D.However, NIKE3D’s requirement for the solution ofcoupled systems of linear equations creates a newand significant challenge. Traditionally, linearequation solvers are divided into two classes:direct and iterative.

Direct methods are known for their robustnessand predictable number of operations. However,they require a global data structure that growsrapidly with problem size and that presents chal-lenges for large-scale parallelism.

Iterative methods typically use a much smallerdata structure that can be readily distributed forlarge-scale parallelism. Unfortunately, the operationcounts for iterative solvers grow enormously in thepresence of ill-conditioning, that is, the existence ofwidely-spaced natural frequencies in the system.Sources of ill-conditioning are most often part of thesimulations we seek to perform: material softeningand damage, material anisotropy, mesh refinementto capture local effects, combined bending andmembrane response of shells, and near-singularitiesas buckling behavior is approached.

These tradeoffs between direct and iterativemethods mean there is no one technology presentlyavailable that totally meet our needs for parallellinear equation solving. Fortunately, this topic is abroad area of inquiry in the academic communityand one that is well supported by multiple effortswithin the ASCI program.

We are monitoring and interacting with theseactivities as we assess current capabilities. As anexample, we completed an interface within NIKE3Dto a parallel direct linear equation solver namedPSPASES from the University of Minnesota.4 Thiswas a prototype effort used to evaluate the linearsolver technology.

The first processor acts as “master”: it performsall I/O, computes the finite element matrices, andthen works in cooperation with the remaining“slave” processors to perform the linear equationsolving. This NIKE3D prototype is running inmessage-passing mode on our DEC cluster and IBMSP (ASCI Blue Pacific).

The test problem examined is the so-calledBoussinesq problem: a concentrated force on thesurface of an elastic half space. In this case thedomain is represented by a mesh of 24 × 24 × 24 hexelements, resulting in a system of 45,000 equations.This highly coupled system of equations is represen-tative of those arising from true 3-D geometries.

The CPU data plotted in Fig. 4 immediately illus-trate one limitation of this particular solver: it onlypermits the use of 2N processors. Comparing the“linear algebra” and “total” CPU times for theuniprocessor run confirms that this simulation isdominated by the cost of linear equation solving.

FY 98 4-49

Figure 2. Arbitrary contact algorithm, used when a sheet ofshell elements is deformed by moving stonewall boundariesthat are the faces of a cube enclosing the sheet

Figure 3. Benchmark test, illustrating the material erosion algorithm.

450 Hoover_qk 7/27/99 7:48 AM Page 4-49


The trends for the parallel runs show reasonablescalability. Given the fixed problem size, the timingcurve will flatten as the number of processorsincreases. The quite mild growth in the gap betweenthe linear algebra CPU time for Processor 0 and thewall time for the factorization seems to indicate thatcommunication latency is not a major penalty evenfor the case of 32 processors.

Finally, for the parallel runs, one sees a larger,though relatively constant offset between the linearalgebra and total CPU time for Processor 0. Thiscontrast from the uniprocessor case reflects the costof mapping from NIKE’s internal representation ofthe global system of equations to that required bythis parallel solver. This cost could be readily elimi-nated by having NIKE assemble into the desired datastructures from the outset.

When assessing the performance of a particularsoftware library, it is important to characterize theeffect of the available algorithmic features and run-time environment.

For example, the data in Fig. 5 illustrate theimpact of two different ordering strategies availablein this package. Sparse direct solvers use a reorder-ing of the equations to reduce the number of non-zero coefficients created during the factorizationprocess. In one case, the “parallel” ordering optionuses multiple processors to collectively perform thisoperation. The “serial” ordering restricts the opera-tion to a single processor.

The latter approach has the advantage that theentire equation structure is simultaneously opti-mized, and thus a better overall ordering is attained.

The developers note that typically the serial optionresults in twenty to thirty percent fewer non-zeroterms in the factorized representation of thematrix, leading to possibly a factor of two reduc-tion in arithmetic operations. These savings arereflected in Fig. 5.

Additional performance improvements will beavailable when we are able to use the SP’s “UserSpace” high-speed communication sub-system. Weare currently restricted to the slower “IP” communi-cation sub-system, due to a combination of limita-tions in this prototype “master-slave” implementa-tion, as well as the vendor’s operating system.

The total memory graph in Fig. 6 illustrates howmemory for this benchmark grows with the numberof processors when using the parallel reorderingoption in PSPASES. To provide a basis for compari-son, note that a uniprocessor execution of NIKE3Dusing a sparse direct solver requires less than600 Mb of total memory.

As mentioned previously, the parallel reorderingcannot reduce factorization fill-in as effectively asserial reordering, thus more memory and floating-point arithmetic operations will be required.However, for sufficiently large problems, the serialreordering will require more memory than is avail-able on a single node of the IBM SP system. In suchinstances the parallel option will be the only viablechoice and we can anticipate the behavior shown inthis graph.

Iterative methods for linear equation solving arenot out of the running, and we look forward to addi-tional improvements through ASCI-related efforts.


Tim

e (s

)

Number of processors ("blue")1 8 16 32

0

100

200

300

700

400

500

600Factorization (wall time)Linear algebra (p0 CPU)Total (p0 CPU)

Figure 4. Timing comparison for the 24 x 24 x 24 Boussinesqproblem of uniprocessor NIKE3D with trials on the IBMR/S6000 SP (ASCI Blue-Pacific), using the PSPASES sparsedirect parallel linear equation solver.

Figure 5. Timing comparison for the 24 x 24 x 24 Boussinesqproblem, using the two reordering strategies available withinthe PSPASES package.

Tim

e (s

)

Number of processors ("blue")1 8 16 32

0

100

200

300

700

400

500

600Factorization (wall time) using parallel orderingLinear algebra (p0 CPU) using parallel orderingTotal (p0 CPU) using serial orderingLinear Algebra (p0 CPU) using serial ordering

450 Hoover_qk 7/27/99 7:48 AM Page 4-50


For example, algorithms such as algebraic multi-grid have proved quite powerful for computationalfluid dynamics and may prove equally useful for solidmechanics. And, looking at the relative strengths ofdirect and iterative methods, it is plausible that theultimate solution will be the blending of the twotechnologies into so-called “hybrid methods.”

One example of this approach is the FETI algo-rithm developed for large-scale aerospace structureby C. Farhat (University of Colorado).5 We arecurrently creating a prototype implementation ofFETI which will be used to evaluate the methodologyon problem classes of interest to LawrenceLivermore National Laboratory (LLNL).

Shared Memory Parallelism

We have been motivated to evaluate sharedmemory parallelism for our finite element programsfor two reasons. Computer companies developingmultiprocessor workstations and workstation clus-ters with up to 32 processors in a box are providingfast, efficiently-coded shared-memory parallel linearsolvers. A second reason is that the target computerarchitecture for the ASCI Blue Mountain and BluePacific computers will consist of multiple processorson shared memory nodes connected with a very highspeed interconnect network.

The ASCI architectures will support hybrid paral-lel models combining shared memory and messagepassing. Although the computer operating systemand compiler software is very immature at thispoint, we have experimented with shared memory

parallel syntax in the DYNA3D program using thenew OpenMP standard.

We had determined in earlier tests with NIKE3Dthat the automated compiler parallelism generatedon a loop level provided parallel granularity that wastoo small to overcome the latency associated withallocating the threads to a processor. The DYNA3Dshared memory implementation uses a large granu-larity for parallel tasks by defining OpenMp parallelregions in the upper level integration routine.

For example, the threads are initialized over thestress divergence calculation by element type.Threads are also distributed across the materials foreach element type. Shared nodal data for the scatterstep in the nodal force update are treated as a criti-cal data region so that the nodal force updateremains sequential. The contact algorithms aretreated by spreading the loops which include largeamounts of computational work over the processors.

The most tedious aspect of this implementationwas the identification of shared and processor-localvariables when parallel threads are initialized. Somemodifications to the COMMON blocks storingelement and material data were required in thisstep. Further modifications would be needed in thefuture for a hybrid model to identify shared andprocessor-local variables which are used in messagepassing data structures. This step would be takenonly if the performance of the shared memory paral-lelism provides large speed gains over messagepassing models with multiple domains on acomputer node.

Our performance results for shared memoryparallelism are still quite preliminary. We find goodperformance for a two-processor SGI Octane work-station and also for some problems on a four-processor SGI Origin 200. On the other hand wehave found it difficult to achieve good performancewith some of our test cases using more than fourprocessors on an SGI Origin 2000. By using perfor-mance analysis tools we were able to identify cachemisses as the most significant limitation on theperformance. Nodal reordering techniques mayprovide a remedy for these performance problemsand may be investigated in the future.

Parallel Smooth-Particle Applied Mechanics

The use of particle techniques to solve the equa-tions of continuum mechanics has become ofincreasing interest over the last decade. This inter-est in particle methods is motivated by the fact thatthey can be used to recast the partial differential

FY 98 4-51

Tota

l mem

ory

(M

b)

Number of processors ("blue" TR)1 8 16 32

0

200

400

600

2000

800

1000

1800

1600

1400

1200

Memory (Mb)

Figure 6. Memory comparison for the 24 x 24 x 24 Boussinesqproblem of uniprocessor NIKE3D with trials, using thePSPASES sparse direct parallel linear equation solver withparallel reordering.

450 Hoover_qk 7/27/99 7:48 AM Page 4-51


equations describing continuum mechanics into aset of ordinary differential equations which may beeasier or more efficient to solve numerically.

Perhaps a more compelling reason for usingparticle methods is that there is no underlying gridin the calculation. Thus, models for fracture, fragmen-tation, penetration mechanics, and plastic flow can bedeveloped without resorting to adaptive grid methods.

The smooth-particle applied mechanics (SPAM)method, often referred to as smooth-particle hydro-dynamics (SPH), is one of several particle methods.

SPAM has been applied to both solids as well asfluid mechanics problems.6,7 This method describesthe continuum as a set of particles with each parti-cle associated with a position and a weighting func-tion. The weighting function has a range that smearsout the particle over a region around its position.The density at any point in space is calculated by thesummed contribution of all particles within therange of the point.

Similarly, other physical quantities such as veloc-ity, strain, stress, and energy are calculated by theweighted averages of the same physical variablesascribed to the particles. Details of this method forconverting the partial differential equations ofcontinuum mechanics into an equivalent set of ordi-nary differential equations are described in theabove references.

Work is in progress to develop a SPAM model foranisotropic granular solid-melt systems generatedduring mold filling and other manufacturingprocesses. Two-dimensional atomistic simulations ofthis system have been completed and will becompared to SPAM calculations for an equivalentcontinuum model.8 A parallel implementation of atwo-dimensional SPAM algorithm is under develop-ment and will be used for simulating solid-meltsystems on ASCI parallel computers.

Parallel Visualization

In anticipation of problem sizes in the range of 15million elements and larger, we are developing aprototype parallel visualization tool based on theGRIZ program. Our initial implementation assumesthe domain decomposition from the finite elementanalysis and uses a master/slave parallel modelwhich splits GRIZ functionality over one masterprocessor and multiple slave processors. The masterprocessor is responsible for all interactive functionsincluding full frame display and interactive menuand mouse commands. The slave processorscompute results and render frames associated withtheir subdomain from the full mesh.

Frame generation proceeds as follows. The masterprocessor receives user input to generate a framedisplaying a result (primal or derived). The masterbroadcasts this command to slave processors thatprocess the command, calculate results, and rendertheir portion of the frame into a local offscreen framebuffer. The slave processors perform a global reduc-tion of the pixel color values, based on their associ-ated z-buffer (depth) values to arrive at a compositemesh image across all subdomains. The compositemesh image is sent to the master, which loads it intothe user’s frame buffer and renders foreground infor-mation such as text, labels, and colormap over it.

We use two MPI (Message Passing Interface)communicator groups to implement this model. Thefirst contains the master and all of the slave nodes.The second contains slave nodes only.

This parallel prototype is implemented in the Miliversion of GRIZ. Mili (Mesh I/O Library) providescode developers with a flexible tool for creating self-defining binary data bases for the visualization ofcomplex grids. The self-defining feature is particu-larly useful for parallel simulations where it isimportant to limit the size of the resulting data sets.

Future Work

The ParaDyn program has matured to a fullproduction status. New algorithms or extensions toexisting DYNA3D algorithms are now being designedwith a plan that includes the parallel implementa-tion. Three projects scheduled for next year are: 1) towrite a users’ manual for the ParaDyn program; 2) toadd message-passing communication to theLagrange contact algorithm; and 3) to complete theprogramming for the few remaining DYNA3D optionsnot implemented in parallel.

Future development projects will include parallelalgorithm development for NIKE3D and a collabora-tive effort with LLNL’s Thermal Fluids Group todevelop parallel versions of TOPAZ3D and fluidcodes used in coupled thermomechanical analysis.

Investigations will continue in grid-free methodssuch as SPAM. These methods may prove to beuseful in capturing highly localized physicalphenomena when used in combination with finiteelement methods which describe a large-scale engi-neering analysis on a parallel computer.

Acknowledgments

We gratefully acknowledge support for ParaDyncode development from the Weapons Program atLLNL and from the Department of Defense High-


450 Hoover_qk 7/27/99 7:48 AM Page 4-52


Performance Computing and ModernizationProgram. We especially thank M. Giltrud from theDefense Threat Reduction Agency (formerly theDefense Special Weapons Agency) for funding bothserial and parallel contact algorithm development.Professor W. Hoover (University of California, Davis)inspired the model for the crumpled material bench-mark for ParaDyn. He has also conducted extensiveresearch on the SPAM algorithm and has developedmany physical applications using the method. Weappreciate the feedback and helpful comments fromanalysts running the ParaDyn program, especiallyD. Badders, T. Lee, and D. Faux from LLNL, andR. Namburu and P. Papados from the U. S. ArmyCorps of Engineers Waterways Experiment Station.

References

1. Hoover, C. G., A. J. De Groot, and R. J. Sherwood(1998), “Parallel Contact Algorithms for ExplicitFinite Element Analysis,” Modeling and SimulationBased Engineering, S. N. Atluri and P. E. O’Donoghue,eds., Tech Science Press, Palmdale, California.

2. Whirley, R. G., and B. E. Engelmann (1993),“DYNA3D: A Nonlinear, Explicit, Three-DimensionalFinite Element Code for Solid and StructuralMechanics—User Manual,” Lawrence LivermoreNational Laboratory, Livermore, California (UCRL-MA-107254, Rev. 1).

3. Benson, D. J., and J. O. Hallquist (1990), “A SingleSurface Contact Algorithm for the Post-BucklingAnalysis of Shell Structures,” Computer Methods inApplied Mechanics and Engineering, Vol. 78, pp. 141-163.

4. Gupta, A., F. Gustavson, M. Joshi, G. Karypis, and V.Kumar (1997), “Design and Implementation of aScalable Parallel Direct Solver for Sparse SymmetricPositive Definite Systems,” Proceedings, Eighth SIAMConference on Parallel Processing.

5. Farhat, C., and F.-X. Roux (1991), “A Method of FiniteElement Tearing and Interconnecting and its ParallelSolution Algorithm,” International Journal for Numerical Methods in Engineering, Vol. 32, pp. 1205-1227.

6. Posch, H. A., W. G. Hoover, and O. Kum (1995),“Steady-state shear flows via nonequilibrium mole-cular dynamics and smooth-particle appliedmechanics,” Physcial Review E, Vol. 52, No. 2, pp. 1711-1720.

7. Hoover, W. G., T. G. Pierce, C. G. Hoover, J. O.Shugart, C. M. Stein, and A. L. Edwards (1994),“Molecular Dynamics, Smoothed-Particle AppliedMechanics, and Irreversibility,” Computers Math.Applic., Vol. 28, No. 10-12, pp. 155-174.

8. Hoover, W. G., and S. Hess (1998,) “AnisotropicPlasticity with Embedded-Atom Potentials,” PhysicaA, in preparation.

FY 98 4-53

450 Hoover_qk 7/27/99 7:48 AM Page 4-53

YNA3D-TOPAZ3D Coupling and DYNA3D-NIKE3D Linkage


Introduction

Over the years, coupled thermal-mechanical engi-neering problems had been solved by a “weakcoupling” approach of mechanics codes and heattransfer codes. The analyst first runs a thermalanalysis for the model by TOPAZ3D,1 a 3-D, implicitheat transfer finite-element (FE) code, and writesthe temperatures at selected times to a file. Thesetemperatures are then read from the file at theircorresponding times in the subsequent mechanicalanalysis, usually by the 3-D, explicitstructural/continuum FE code DYNA3D.2 Thecoupling is achieved through the use of temperature-dependent material properties or temperaturechange-induced thermal stresses. From the struc-tural analysis standpoint, this procedure does reflectthe temperature-sensitive nature of the problem, butdoes not quite capture the essence of thermal-mechanical interaction.

With the evolution of the weapon systems and theexpanding scope of laser-related applications atLawrence Livermore National Laboratory, theimportance of detailed thermal-mechanical stressanalysis has become evident. The integration ofDYNA3D and TOPAZ3D is intended to accomplishthis goal. When requested by a user, TOPAZ3Dwould be compiled as a program module inDYNA3D. For a thermal-mechanical analysis, thecoupling is then achieved by the continuousexchange of deformation and temperature databetween these codes on a real-time basis.

The link between DYNA3D and NIKE3D,3 animplicit, structural/continuum FE code, is arrangedin a serial mode and intended to solve certainclasses of structural/continuum mechanics problem.This combined explicit-implicit method is particu-larly effective for simulations involving initial stress,such as failure analysis of rotating fan blades, orresidual stress, such as sheet metal forming processwith spring-back.

Due to its explicit nature, DYNA3D would havedifficulty attaining an equilibrium state for therotation-induced initial stresses in the fan blades orsustaining a long duration simulation of spring-backphenomena, whereas the implicit NIKE3D isperfectly suited for these purposes. However, theimpulsive loading, high nonlinearities and compli-cated contact conditions in the transient phase ofthese problems usually render NIKE3D ineffectiveand necessitate the use of DYNA3D.

The capability of reading a file generated by itsopposite code, and writing a file for its oppositecode is implemented in both DYNA3D and NIKE3D.This binary file contains the nodal kinematic vari-ables and element stresses, strains, and other statevariables. Series of DYNA3D and NIKE3D analyses,with essential data transmitted through this file, canbe strung along to solve complicated multi-phaseproblems. Different element formulations arepreferred in DYNA3D and NIKE3D because of theirexplicit/implicit natures. Change of element formula-tion is permitted during the transition from one codeto the other to accommodate this fact.

FY 98 4-55

This report describes the coupling of DYNA3D and TOPAZ3D and the linkage between DYNA3D andNIKE3D. The combined DYNA3D-TOPAZ3D code provides a tool for coupled thermal-mechanicalanalyses, and the DYNA3D-NIKE3D link enables the multi-staged simulations for structural/continuummechanics problems. For a coupled thermal-mechanical analysis, DYNA3D and TOPAZ3D run inparallel and exchange structure/continuum deformation and temperature as the analysis progresses.For structural/continuum mechanics problems involving different regimes of loading, deformation, orfrequency characteristics, the DYNA3D-NIKE3D link is designed to run in series. It aims to fully usethe explicit/implicit advantages of the individual codes at the proper stages.

Jerry I. LinDefense Technologies Engineering DivisionMechanical Engineering

455 Lin_qk 7/27/99 7:50 AM Page 4-55


Progress

DYNA3D-TOPAZ3D Coupling

In DYNA3D, temperature dependency can beestablished through the use of material type 4(thermal-elastic-plastic), type 15 (Johnson/Cookelastic-plastic), type 21 (thermal-orthotropic-elastic)or type 23 (thermal-orthotropic-elastic with variableproperties). They have either the material temperatureas part of the constitutive equations, or temperature-dependent material properties. Because of differenttime increments used in time integration in thesecodes—usually a greater time step in the implicitTOPAZ3D, and a smaller one in the explicitDYNA3D—the coupling can be executed only whenthe simulation times in both codes are synchronized.The temperatures at the intermediate DYNA3D stepsare obtained by linear interpolation.

The coupling procedure, depicted in Fig. 1, canbe described as follows:

1. DYNA3D feeds current nodal coordinates toTOPAZ3D when simulation clocks in bothcodes are synchronized.

2. TOPAZ3D does a thermal analysis and returnsthe updated current temperature to DYNA3Dat the end of this thermal step. TOPAZ3Dadvances its clock.

3. DYNA3D executes the mechanical analysisbased on the current temperatures provided byTOPAZ3D and advances its clock incrementallyuntil the clock catches the TOPAZ3D clock.

This process repeats itself until it reaches thedesignated simulation time.

A TOPAZ3D compilation flag is set in DYNA3D.Only if the flag is activated will TOPAZ3D be inte-grated into DYNA3D. A special arrangement is alsomade to have the temperatures written into theDYNA3D plot database for post-processing purposes.

DYNA3D-NIKE3D Link

Due to the explicit nature of DYNA3D, theBelytschko-Lin-Tsay shell element4 and theuniformly under-integrated hexahedral elements(URI)5 are the preferred elements in DYNA3D. Onthe other hand, the Hughes-Liu shell element andthe Selective-Reduced Integrated (SRI)6 hexahedralelement are favored by NIKE3D because of itsimplicit integration approach.

To take full advantage of the codes’ characteris-tics, different elements can be used in these codes.A transformation of the shell element stresses to theglobal coordinate system is performed beforeDYNA3D writes them into the link file. It ensures thestresses NIKE3D reads are as expected by the code.A reverse stress transformation is also done rightafter DYNA3D reads the link file generated byNIKE3D. For hexahedral continuum elements,NIKE3D averages the stresses at element integra-tion points to give DYNA3D elements a uniformstress state, and imposes uniform stresses onelement integration points while reading thestresses from the DYNA3D-generated link file.

The SRI hexahedral elements as well as a fullyintegrated shell element7 have been implemented intoDYNA3D, whereas the URI and shell elements havebeen added to NIKE3D, to give both codes a completeelement library. In the event that high fidelity transferfrom one code to the other is necessary, users havethe choice to use unified element formulations, albeitit may come with a high price on DYNA3D analysis.

Figure 2 depicts the sequence for a typical multi-phase mechanical analysis. A Pokel Cell example(Fig. 3) is used to demonstrate the advantages ofthe combined analysis. In this problem, an O-ring isbeing pressed into the seal groove by a piece oflaser glass. Because of the O-ring’s viscous elasticmaterial behavior, the response includes a sharprise of stress in the loading phase and a prolongedrelaxation in the unloading phase. DYNA3D canhandle the transient loading phase with ease, butwould have difficulty bringing the stress to a relaxedsteady state. The use of NIKE3D during the unload-ing phase results in a satisfactory solution, asshown in the stress-time plot in Fig. 3.


DYNA3D

TOPAZ3D

Time

Simulation clock advance

Temperature input flow (TOPAZ3D to DYNA3D)Deformation input flow (DYNA3D to TOPAZ3D)Clock synchronization

Figure 1. Illustration of DYNA3D-TOPAZ3D coupling procedure.

Stressinitialization

Transientanalysis

Residual stressproblem

NIKE3D DYNA3D NIKE3D

Figure 2. Illustration of DYNA3D-NIKE3D link sequence fortypical multi-phase mechanical analysis.

455 Lin_qk 7/27/99 7:50 AM Page 4-56


In large scale models, especially with compli-cated contact conditions or a long simulation time,DYNA3D users sometimes encounter spatial insta-bility that cannot be controlled by the existing anti-hourglass algorithms. With the multi-point inte-grated elements now available in DYNA3D, usershave the option to deploy them at critical parts ofthe mesh that are particularly susceptible to hour-glass modes.

Future Work

For coupled thermal-mechanical analyses, thepossible temperature change due to the dissipationof element plastic strain energy still needs to beaccounted for. This plastic strain energy must becomputed by DYNA3D and transferred along withthe deformation to TOPAZ3D.

On the DYNA3D-NIKE3D link, we are lookinginto the implementation of global stress interpola-tion schemes in DYNA3D. It would enableDYNA3D to provide non-uniform element stresses,without the high cost of multi-point integratedelements, to NIKE3D.

Acknowledgments

The DYNA3D-TOPAZ3D coupling is a joint effortby A. Shapiro and the author. The DYNA3D-NIKE3Dlink and the implementation of new elements are ateam effort by M. Puso, E. Zywicz, and the author.

References

1. Shapiro, A. B. (1985), TOPAZ3D—A Three-Dimensional Finite Element Heat Transfer Code,University of California, Lawrence Livermore NationalLaboratory, Livermore, California UCID-20484.

2. Whirley, R. G., and B. E. Engelmann (1993),DYNA3D: A Nonlinear, Explicit, Three-DimensionalFinite Element Code for Solid and StructuralMechanics—User Manual, University of California,Lawrence Livermore National Laboratory, Livermore,California UCRL-MA-107254.

3. Maker, B. N. (1995), NIKE3D: A Nonlinear, Implicit,Three-Dimensional Finite Element Code for Solid andStructural Mechanics—User’s Manual, University ofCalifornia, Lawrence Livermore National Laboratory,Livermore, California UCRL-MA-105268.

4. Belytschko, T., J. I. Lin, and C. S. Tsay (1984),“Explicit Algorithms for the Nonlinear Dynamics ofShells,” Computer Methods in Applied Mechanics andEngineering, 42, pp. 225-251.

5. Goudreau, G. L., and J. O. Hallquist (1982), “RecentDevelopments in Large Scale Finite ElementLagrangian Hydrocode Technology,” ComputerMethods in Applied Mechanics and Engineering, 30.

6. Hughes, T. J. R. (1987), The Finite Element Method:Linear Static and Dynamic Finite Element Analysis,Prentice-Hall, Princeton, New Jersey, pp. 232-237.

7. Bathe, K., and E. N. Dvorkin (1985), “A Four-NodePlate Bending Element Based On Mindlin/ReissnerPlate Theory and a Mixed Interpolation,”International Journal for Numerical Methods inEngineering, 21, pp. 367-383.

FY 98 4-57

Effe

ctiv

e st

ress

57.350.040.030.020.010.0

1.0

Effective stress

3.6

3.4

3.2

3.0

2.8

2.6

2.4

2.2

2.0

1.8

1.6-4 -2 -1-3 0 1 2

Time

DYNA3Danalysis NIKE3D analysis

Figure 3. DYNA3D-NIKE3D example: stress-time plot, showinguse of NIKE3D during the unloading phase.

455 Lin_qk 7/27/99 7:50 AM Page 4-57

Physically-Stabilized Eight-Node Hexahedral Element


Introduction

The explicit and implicit finite-element methodshave spawned two different element types to maxi-mize their efficiencies. The bottleneck for theexplicit method is the strain and constitutive evalua-tions. Consequently, we have used a single-pointintegration method, where these evaluations aremade only at the center of the element. To make upfor the loss of resolution from the single-point inte-gration, more elements are used in the calculation.

On the other hand, the bottleneck for the implicitmethod is the inversion of a system of linear equa-tions. Consequently, it is desirable to reduce thenumber of equations/elements in the analysis byusing a higher-order integration rule over the spatialdomain of the element, to gain resolution andstability. However, the increased use of mixedimplicit/explicit analysis necessitates a new elementstrategy that combines the austerity of the single-point integration with the robustness of the multiple-point integration. In this work, a novel single-pointintegration scheme is developed that performs aswell as the fully integrated element for many prob-lems but is much more efficient computationally.

Progress

The single-point integration method requireshourglass control to eliminate zero-energy modes, toremain stable. The perturbation method currentlyused in explicit finite-element codes requires ad hocparameters to calculate the hourglass forces and,consequently, can give poor results particularly forcoarse meshes.

In this work, we use the physical stabilizationmethod to compute the hourglass forces by a Taylorseries expansion of the stresses and strains aboutthe center of the element. This allows the element tobe integrated in closed form, providing the exacthourglass forces for initially parallelepipedelements. Consequently, no artificial parameters arerequired to preclude hourglassing.

As many analysts know, some problems requirecalibration of hourglass parameters to eliminatehourglassing in a mesh while avoiding spurious stiff-ening of the structure. This new method should elim-inate this time-consuming procedure and providebetter coarse mesh accuracy, particularly forbending-dominated problems.

After contact, the hourglass control forces aretypically the most expensive calculation in theexplicit method. For an all brick element problemwith no contact and a simple elastic material,stiffness-based hourglass forces comprise twentypercent of the total CPU time. With the new physicalstabilization method, the hourglass forces willcomprise about thirty percent of the total CPU time.Overall, this amounts to a ten percent increase intotal CPU time for the worst case.

With problems including contact and more expen-sive material models, the extra time for the physicalstabilization will be further amortized. Compare thisto a fully integrated element, which takes betweenfour and six times more CPU time in the elementcalculation than the perturbation method.

This physical stabilization method is implementedinto DYNA3D and NIKE3D for general use and canbe exploited in mixed explicit/implicit analysis viathe DYNA3D/NIKE3D link, or the NIKE3Dimplicit/explicit analysis mode.

FY 98 4-59

We have developed a novel single-point integration scheme that performs as well as the fully inte-grated element for many problems, but is much more efficient computationally. We used the physicalstabilization method to compute the hourglass forces by a Taylor series expansion of the stresses andstrains about the center of the element.

Michael A. PusoDefense Technologies Engineering DivisionMechanical Engineering

460 Puso_qk 7/27/99 7:52 AM Page 4-59


Theory

The perturbation hourglass control method canapply either elastic or damping forces to resist thehourglass modes. Although slightly more efficient,the damping-based method is entirely non-physicaland is not applicable to static problems, sincemotion is required to resist the modes. Furthermore,with the damping method, hourglass modes willalways eventually become evident in problemswhere boundary conditions allow them to propagate.Only the stiffness-based hourglass control methodsare considered here.

Flanagan and Belytschko1 were the first to usethe orthogonal hourglass control method for 3-Dnonlinear transient analysis. This method uses theso called γ stabilization vectors, which are orthogo-nal to the homogenous strain states, and conse-quently satisfy the patch test, providing a convergentformulation for hourglass control. In the firstappearance of physical stabilization,2 this methodwas applied to shells and four-node quadrilateralelements (2-D).

The first physical stabilization method applied toeight-node hexahedral elements (3-D) was byBelytschko and Bindeman.3 This method alleviatedthe problems of locking due to incompressibility,and avoided the parasitic shear locking in bending-dominated problems for undistorted elements. Butthe element required a co-rotational coordinatesystem, which is only well-posed for parallelepipedelements and in general is not indifferent toelement node number.4

Furthermore, hourglass forces are not convectedwith the deformation in a consistent fashion. Zhuand Cescotto5 proposed a general method for physi-cal stabilization of the eight-node hexahedral butrequired the storage of 36 hourglass stresses, asopposed to the twelve hourglass forces quantitiesusually needed. Also, no valid solution to parasiticshear locking was offered.

In this work we formulate a physical stabilizationmethod that uses the convected frame in lieu of aco-rotational frame.3 This allows us to use anassumed strain field for the hourglass forces, whichcan avoid shear lock yet be frame-indifferent.Furthermore, only twelve hourglass forces areneeded, and hourglass forces are correctlyconvected for material models using the so-calledconvected or Truesdale rate (for example hyperelas-ticity, Fe/Fp plasticity model).

For ordinary general hypoelastic models thehourglass forces are valid for small strain/large rota-tion, but should still be effective.

In the finite-element method, the internal forcesfor an element with domain Ωe is given by

, (1)

where B is the strain displacement matrix and σ isthe stress. In single-point integration, only thevalues of the stress and strain displacement at theelement center (σo and Bo) are used in Eq. 1 so thatthe internal force is calculated at time n+1 by

. (2)

The additional forces used to preclude the zero-energy modes are given by the update:

,

(3)

where

,

, (4)

γi is the ith 8 x 1 stabilization vector, ∆u is the incre-mental displacement, Q is the incremental rotationgiven by the spin at the element center, fi is the ith

3 × 1 hourglass force, and α is the artificial hour-glass control parameter.

In the physical stabilization method, it is recog-nized that the strain matrix has a constant and bilin-ear (or hourglass) portion, such that B = Bo + Bhg.Furthermore, by using the following approximation:

, (5)

to calculate the hourglass portion of the straindisplacement matrix, certain orthogonality condi-tions can be exploited such that Eq. 1 can be inte-grated in closed form.

In this work the Truesdale rate is used as theobjective rate in the constitutive law, such that thevolume integration can precede the constitutivetime integration in Eq. 1, and the 36 hourglassstresses per element needed (that is, stored) arecondensed to just 4 3 × 1 hourglass forces neededin the update.

∂∂

∂ξ∂

∂∂ξξ

( )

( )⋅ ≈ ⋅

=x x 0

Γi

i

i

i

=

γγ

γ

0 0

0 0

0 0

k Iij ij = ×α δ 3 3

f k u Qfin

ij jT

in+ = +1 Γ ∆

F fhgn

i in

i

+ +

=

= ∑1 1

1

4

Γ

F Bint

noT

o+1 = ( ) ( )+ +∫ t t dVn n

e1 1σ

Ω

F Bint = ∫ T dV

e

σΩ


460 Puso_qk 7/27/99 7:52 AM Page 4-60


An assumed strain field is used for thecontravariant strain tensor so that parasiticshear lock is avoided for bending-dominated prob-lems. The following form is used to calculate thehourglass forces:

,

(6)

where

, (7)

Jo is the 3-x-3-element Jacobian at the elementcenter, C is the 6-x-6 material stiffness, J is the6-x-6 matrix that transforms the incremental strainvector onto the covariant coordinate system form bythe columns in Jo, and Bij is a 6-x-3 matrix thataccounts for the individual contravariant strains.

Although Eq. 7 looks ominous, about half the Bijsare zero matrices. The other half are sparse, in thatthey include at most two ones. The remaining termsare zeros. Furthermore, some redundancies can beexploited in the implementation. The bottom line isthat Eq. 7, when implemented appropriately, is notexorbitantly expensive. As mentioned, the newmethod takes about 50% longer than the perturba-tion method, resulting in at most a ten percentincrease in total CPU time.

Examples

Many problems work quite well with the pertur-bation method hourglass control, depending on theboundary conditions and material propertiesinvolved. In problems with highly anisotropic mater-ial properties or point loads that occur due toboundary conditions or contact, the perturbationmethod may not eliminate all hourglassing and givepoor results in coarse meshes. Furthermore, hour-glass parameters that are too big cause the problemto be overly stiff.

The following plate problem illustrates thepenalty sensitivity that can sometimes occur. Asquare plate with the following material properties:

Ex = 50000 ksi, Ey = 100 ksi, Ez = 100 ksiGx = 10000 ksi, Gx = 10000 ksi, Gx = 10000 ksiVyx = 0.00075, vzx = 0.3, vzy = 0.3

k J B B Jij o il

T Tjl o

T

l

– –==∑ J C J 1

1

6

f k u J fin

ij jT

o in+ = +1 Γ ∆

F fhg

ni i

n

i

+ +

=

= ∑1 1

1

4

Γ

is loaded at the center, as shown in Fig. 1. Thehourglass parameter α from Eq. 3 is chosen to besome percentage of the maximum Eigenvalue in thestiffness matrix. In DYNA3D and NIKE3D thisamounts to the following equation for theorthotropic material models:

, (8)

where Bo is the strain displacement matrix and κ = 0.1 by default. This value of κ proves overlystiff, and three other values for κ are tried: κ = 0.00002, 0.0002, and 0.002. The value κ = 0.1 (100/50000) = 0.0002 replaces Emax withEmin in Eq. 8 and uses the default 0.1, whichseems to work well with isotropic materials.

Two different loadings shown in Figs. 1 and 2 areanalyzed with the five different meshes (Table 1).

Figures 3 and 4 show results for the differentmeshes using the different κs, the physical stabi-lization method (denoted by ps) and full integration(using the incompatible modes element). For theload case in Fig. 1, from Fig. 3 it is seen that thephysical stabilization method converges at aboutthe same rate as the fully integrated element. Withκ = 0.0002, the perturbation method performswell, whereas with 0.00002 it is too soft, and with0.002 it is too stiff for the mesh densities used.

α κ . max= ⋅ ⋅ ⋅ ( ) ( )0 05 E o ij o ij

B B

FY 98 4-61

Table 1. Matrix parameters.

nz x nx x ny No. of elements

1 × 6 × 6 363 × 6 × 6 543 × 13 × 13 5076 × 26 × 26 40567 × 38 × 38 10108

F F

Z Y

X

Figure 1. Loading case 1: 20-in.-x-20-in.-x-1-in. plate with8-lb edge loads and simple supports.

460 Puso_qk 7/27/99 7:52 AM Page 4-61


For the load case in Fig. 2, we get some hour-glassing for κ = 0.00002 and 0.0002, as seen inFig. 5. For κ = 0.002, we get no hour glassing, butas seen in Fig. 4, the perturbation method gaveresults that were too soft. The physical stabiliza-tion method gives no hourglassing (Fig. 6), andagain converges at the same rate as the fully inte-grated element.

The conclusion to be drawn from these examplesis that the perturbation method does not performwell under certain types of loadings and the resultsare highly sensitive to the values of the stiffnessparameter, κ. The new physical stabilization methodis very reliable since its hourglass forces are derivedfrom the fully integrated element.


0

0.1

0.2

0.3

0.4

0.5

1000100

0.002

0.0002

0.00002

Full

ps

Max

imum

dis

pla

cem

ent

(in

.)

Number of elements

Exactsolution 0.164 in.

Figure 3.Convergence plot forloading case 1.

1000100 104

Max

imum

dis

pla

cem

ent

(in

.)

0

0.5

1.0

1.5

2.0

2.5

Number of elements

Exactsolution1.65 in.

0.002

0.0002

0.00002

Full

ps

Figure 4.Convergence plot forloading case 2.

F

F

F

F

Figure 2. Loading case 2: 20-in.-x-20-in.-x-1-in. plate with32000-lb edge loads and clamped supports.

460 Puso_qk 7/27/99 7:52 AM Page 4-62


Future Work

The physically-stabilized element works well,particularly for elasticity. Even with plasticity, thephysical stabilization method offers a consistentmethod, capturing plasticity in the hourglass modes.Nevertheless, with plasticity the element is either allplastic or elastic because of the single-point integra-tion. The fully integrated element, on the other hand,can resolve the plastic front more accurately incoarse meshes. This partial plasticity could poten-tially be incorporated into the hourglass force calcu-lation by monitoring the hourglass strains and modi-fying the hourglass stiffness appropriately.

Although the assumed strain field used in thephysically-stabilized element works well for manyproblems, like all assumed-strain methods wheresome portion of the strain field is projected out, theresults can sometimes be diffusive. Higher-orderenhanced strain methods could be used in a physicalstabilization setting to give better resolution.

References

1. Flanagan, D. T., and T. Belytschko (1981), “A UniformStrain Hexahedron and Quadrilateral with OrthogonalHourglass Control,” IJNME, 17.

2. Liu, W. K., J. S. Ong, and R. A. Uras (1985) “FiniteElement Stabilzation Matrices—A UnificationApproach,” CMAME, 53.

3. Belyschko, T., and L. P. Bindeman (1993), “AssumedStrain Stabilization of the Eight-Node HexadralElement,” CMAME, 105.

4. Crisfield, M. A., and G. F. Moita (1996), “A Co-Rotational Formulation for 2-D Continua IncludingIncompatible Modes,” IJNME, 39.

5. Zhu, Y. Y., and Cescotto, S. (1996), “Unified and MixedFormulation of the Eight-Node Hexahedral Elementsby Assumed Strain Method,” CMAME, 129.

FY 98 4-63

Figure 5. Loading case 2 with κ = 0.00002. Figure 6. Loading case 2 with physical stabilization.

460 Puso_qk 7/27/99 7:52 AM Page 4-63

Cyclic Viscoplastic Constitutive Model


Introduction

The life prediction of structures and structuralcomponents subjected to complex loading histories,which include repeated loading and unloading,requires an accurate representation of the materialbehavior. Typically, at the very least, this includes anaccurate representation of monotonic and cyclichardening and softening behavior, transient andstabilized hysteresis behavior, and the non-fadingmemory effects of the material. However, thematerial models that are currently available inNIKE3D do not have the capability to represent theabove-mentioned material behavior.

Furthermore, the implementation of many ofthese material models into NIKE3D is intrinsicallylinked to the element technology. That is, the samematerial model is implemented separately forelements such as bricks, shells, and beams. Thisinconvenience is quite apparent when the user isforced to use a certain type of element not becauseof the geometry and loading conditions of the prob-lem, but because of the availability of the requiredmaterial model to that certain element type.

In this work, a unified cyclic viscoplastic constitu-tive model along with the multi-component forms ofnonlinear kinematic and isotropic hardening rules, isimplemented for an accurate prediction of thecomplex cyclic structural response.1–3

Armstrong-Frederick type rules are used todescribe the nonlinear evolution of each of themulti-component kinematic hardening variables. Asaturation type (exponential) rule is used todescribe the evolution of each of the isotropic hard-ening variables. The concept of memory surface isused to describe the strain-range-dependentmaterial memory effects that are induced by theprior strain histories.1,4

In addition, the algorithmic treatment of theabove constitutive model is such that it allows asingle implementation for any desired stress- orstrain-constrained subspace. Hence, the cyclicviscoplastic model can be used with any of theelements currently available in NIKE3D.

Progress

A cyclic viscoplastic constitutive model thatincludes material memory effects has been imple-mented into NIKE3D. In the following, a briefdescription of the constitutive model is presented inthe standard generalized materials framework.1–3,5

For the sake of simplicity of notation in theconstrained subspaces (such as plane stress, andbeam and shell kinematics), we represent the tenso-rial variables in terms of their corresponding matrixand vector forms.

FY 98 4-65

We have implemented a cyclic viscoplastic constitutive model that includes material memoryeffects into NIKE3D. The constitutive model is based on multi-component forms of kinematic andisotropic hardening variables in conjunction with von Mises yield criteria. In addition, a memorysurface (non-hardening surface) is used to capture the strain-range-dependent material memoryeffects. This work addresses the issues involved in the complete algorithmic treatment of the rate-dependent constitutive model for any desired stress- or strain-constrained configuration subspace.That is, beam, shell, plane stress, plane strain and other stress- and strain-based kinematicconstraints can be handled within a single framework. The constitutive model is capable of represent-ing the cyclic hardening and softening behavior, transient and stabilized hysteresis behavior, and thenon-fading memory effects of the material.

Phani Kumar V. V. NukalaNew Technologies Engineering DivisionMechanical Engineering

TA465 Nukala_qk 7/27/99 7:57 AM Page 4-65


In the standard generalized materials framework,the cyclic viscoplastic constitutive equations areexpressed as:

• additive decomposition of total strain, ε, intoelastic (ε e) and plastic (ε p) parts:

(1)

• state laws based on the form of Helmholtzenergy Ψ:

(2)

(3)

(4)

• evolution laws of internal variables (based onplastic flow potential):

(5)

(6)

(7)

• yield function f(σ, X, R):

, (8)

where

(9)

(10)

• loading/unloading and consistency conditions(rate-independent plasticity):

(11)

• viscoplasticity (Perzyna rule):

f f f ; ˙ ; ˙ ; ˙ ˙ ≤ ≥ ≡ =0 0 0 0λ λ λ

R Rii

= ∑

X X ii

= ∑κ σ = +0 R

η σ – = ( )X

f PT – = ≤1

213

02η η κ

˙ ˙ – r

RQi

i=

23

1 1κλ

α i =λP η−2

3κai Xi

˙ ˙ε λ ηp P=

R

rb Q ri

ii i i = =∂

∂Ψ

X C Ri

ii i := =∂

∂ααΨ 2

3

σ ∂

∂εε := =Ψ

eeC

ε ε ε = +e p

, (12)

where C denotes the elastic constitutive matrix inthe constrained subspace, α i denotes the vectorform of the tensor variable associated with the kine-matic hardening, and ri denotes a scalar variableassociated with the isotropic hardening.

In addition, σ represents the Cauchy stress tensorand Xi and Ri represent the thermodynamic forcesassociated with αi and ri, respectively. The mapping,P, maps the symmetric rank-two stress tensors inthe constrained-stress subspace onto symmetricrank-two strain tensors in the deviatoric subspace.

In the following, a non-hardening memory surfaceis introduced to model the non-fading memoryeffects of the material.

Let g represent the memory surface, defined as

, (13)

where the reduced plastic strain, ζ, is defined asζ = εp – β. The evolution of the memory surface isdefined through the following relations:

(14)

.(15)

The Kuhn-Tucker optimality conditions along withthe consistency requirement can be stated as

. (16)

The main difference between the rate-dependentand rate-independent cases is in the evaluation ofthe plastic multiplier, . The other conceptualdifference is that, in the case of rate-independentplasticity, f ≤ 0, whereas in the case of viscoplasticity,f > 0. However, the numerical treatment for rate-independent and rate-dependent cases is similar.5,6

A generalized midpoint algorithm is used to inte-grate the rate-constitutive equations. Furthermore, theconsistent tangent operator is obtained through anexact linearization of the return mapping algorithm.5

Numerical Examples

The return mapping algorithm and the corre-sponding consistent tangent operator are imple-mented into NIKE3D.7 In the following, several

λ

g g g ; ˙ ; ˙ ; ˙ ˙ ≤ ≥ ≡ =0 0 0 0µ µ µ

˙ ˙ ˙ ˙ ˙q R R gT T=

=

=23

23

12

12

β β µ ζ ζ µ

˙ ˙β µζ=

g R qT – =

=23

02ζ ζ

˙ λ = fK

n




numerical examples representing the cyclic harden-ing and softening behavior, transient hysteresisloops, the non-fading memory effects and rate-dependent effects of the material are presented.These examples demonstrate the accuracy androbustness of the present algorithmic framework.

Uniaxial Stress Response for Aluminum AlloyAA6060: Multiple-Step Test. In the following,numerical simulation of a multiple-step test8 witheight steps of increasing strain amplitude isperformed. This example demonstrates plasticstrain-range-dependent hardening (memory) effects

on the cyclic stress response. Ten cycles areperformed during each step. The material modelconsists of two independent kinematic hardeningvariables and two independent isotropic hardeningvariables. Memory effects include both plasticstrain-range-dependent isotropic and kinematichardening variables. The material parameters foraluminum alloy AA6060 (temper T4) are given inTable 1.

Figures 1a and 1b show the stress response andthe hysteresis loops obtained for the multiple-steptest during increasing and decreasing strain

FY 98 4-67

Table 1. Material parameters for aluminum alloy AA6060.

E = 66240 MPa v = 0.3 σ = 42 MPaQx1= 28 MPa DM1 = 800 D01 = 4800 δ 1 = 400Qx2= 38 MPa DM2 = 12 D02 = 72 δ 2 = 400b1 = 25 QM1 = 75 MPa Q01 = 22.5 MPa ω1 = 80b2 = 0.5 QM1 = 65 MPa Q02 = 19.5 MPa ω2 = 80

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2-250

-200

-150

-100

-50

0

50

100

150

200

250

Strain (%)

Stre

ss (

MPa

)

(a)

-2 -1.5 -1 -0.5 0 0.5 1 1.5 2-250

-200

-150

-100

-50

0

50

100

150

200

250

Strain (%)

Stre

ss (

MPa

)

(b)

Figure 1. Symmetric multiple step test: (a) increasing strainamplitudes; (b) decreasing strain amplitudes.

Figure 2. Variable amplitude test: (a) type 1; (b) type 2.

-1.5 -1 -0.5 0 0.5 1 1.5-250

-200

-150

-100

-50

0

50

100

150

200

250

Strain (%)

Stre

ss (

MPa

)

(a)

-1.5 -1 -0.5 0 0.5 1 1.5-250

-200

-150

-100

-50

0

50

100

150

200

250

Strain (%)

Stre

ss (

MPa

)

(b)



amplitudes, respectively. From the numerical simu-lation response, it should be noted that cyclic hard-ening occurs not only during the progressive cycleswithin a step but also during the increase in step(plastic strain range) size.

Uniaxial Stress Response for Aluminum AlloyAA6060: Variable Amplitude Test. This test8

demonstrates the influence of increasing and

decreasing strain amplitudes on the cyclic stressresponse. The first test is conducted in three stepswith strain amplitudes 0.8, 0.4 and 1.0%. In thesecond test, three steps with strain amplitudes 0.4,0.6 and 0.8% are used. Five cycles are performedduring each step. The material parameters used inthe numerical simulation are identical to those usedin the multiple-step test. The stress response


Figure 3. Uniaxial tensile stress-strain behavior: (a) effect of strain rate; (b) effect of strain rate jumps; (c) harding-relaxation stress-strain curve; (d) harding-relaxation stress versus time.

0 0.5 1 1.5 2 2.5 3 3.5 40

50

100

150

200

250

(a) Strain rate = 5 x 10-4

(b) Strain rate = 1 x 10-4

(c) Strain rate = 1 x 10-5

(d) Strain rate = 2 x 10-8

(e) Strain rate = 1 x 10-12

(a)(b) (c)

(d)(e)

Strain

(a)

0 0.002 0.004 0.006 0.008 0.010

50

100

150

200

250

300

Strain

Stre

ss (

MPa

)

(b)

0 1 2 3 4 5 60

50

100

150

200

250

Strain

Strain rate = 10-4

(c)

0 500 1000 1500 2000 2500 3000 3500 40000

50

100

150

200

250

Time (s)

Stre

ss (

MPa

)

(d)

x10–3

x 10–3

Table 2. Material parameters for rate-dependent effects.

E = 185000 MPa v = 0.3 σ 0 = 82 MPaK = 151 MPa n = 24D1 = 4200 Qx1 = 38.667 MPaD2 = 37.5 Qx2 = 180 MPab1 = 8 Q1 = 60 MPa



obtained for the first and second tests are shown inFigs. 2a and 2b, respectively. For both sequences ofvariable strain amplitudes, the stress responseobtained from the numerical simulations is in goodagreement with the physical behavior of the material.

Rate-Dependent Effects. In the following, theeffect of strain rate on the stress response is illus-trated through three examples. The material modelconsists of two independent kinematic hardeningvariables and an isotropic hardening variable todescribe the material behavior. A power law is usedto describe the rate-dependent effects. The materialparameters used for the problems concerning rate-dependent effects are given in Table 2.

Figure 3a shows the effect of strain rate on themonotonic tensile behavior of 316L stainless steel.The influence of change in rate of strain during thetest on the monotonic stress-strain curve is shownin Fig. 3b. From Fig. 3b, it can be seen clearly thata change in the rate of strain during the test resultsin an immediate change in the stress-strain curve,and tends to rejoin the monotonic stress-straincurve corresponding to the new strain rate. Figures3c and 3d represent the numerical simulation ofhardening-relaxation behavior of the material.

Future Work

Modeling of ratcheting effect, static recoveryeffects, and the effect of temperature on thematerial behavior is the scope of the future work.Formulation and a complete algorithmic treatmentof the above cyclic viscoplastic constitutive model inthe finite strain hyper-elastic-plastic constitutiveframework are also planned.

References

1. Lemaitre, J., and J. L. Chaboche (1990), Mechanicsof Solid Materials, Cambridge University Press,Cambridge, UK.

2. Chaboche, J. L. (1989), “Constitutive equations forcyclic plasticity and cyclic viscoplasticity,”International Journal of Plasticity, 5, pp. 247–302.

3. Ohno, N. (1990), “Recent topics in constitutivemodeling of cyclic plasticity and viscoplasticity,”Applied Mechanics Review, 43, p. 283.

4. Ohno, N. (1982), “A constitutive model of cyclic plas-ticity with a non-hardening strain region,” Journal ofApplied Mechanics, 49, p. 721.

5. Nukala, P. K. V. V., “A Cyclic Viscoplastic ConstitutiveModel,” submitted to Computer Methods in AppliedMechanics and Engineering.

6. Simo, J. C., and S. Govindjee (1991), “Non-linear B-stability and symmetry preserving return mappingalgorithms for plasticity and viscoplasticity,”International Journal of Numerical Methods inEngineering, 31, pp. 151–176.

7. Maker, B. N. (1995), “NIKE3D: A nonlinear implicitthree-dimensional finite element code for solid andstructural mechanics,” Lawrence Livermore NationalLaboratory, Livermore, California (UCRL-MA-105268,Rev. 1).

8. Langseth, M., O. S. Hopperstad, and S. Remseth(1995), “Cyclic stress-strain behavior of alloyAA6060, Part I: Uniaxial experiments and model-ing,” International Journal of Plasticity, 11, pp. 725–739.

FY 98 4-69


lectromagnetic Cold-Test Characterization of the Quad-Driven Stripline Kicker


Introduction

The original kicker design1-3 was conceived toallow for the diversion of the electron beamdynamically during a long pulse, thus acting like abeam-splitter (Figs. 1 and 2). Experimentsperformed on the kicker4 detail the operatingparameters of the system. This report outlines theelectromagnetic cold-test measurements

performed on the kicker as part of the analysis,and concepts for the kicker pulser requirements.

Due to beamline use and the motivations for thecold-test measurements, the kicker was tested inthe Lawrence Livermore National Laboratory(LLNL) Electromagnetics Laboratory using a vari-ety of vector network analyzers to sweep thefrequency band, and time-domain impulse genera-tors and scopes.

FY 98 4-71

The first kicker concept design for beam deflection was constructed to allow stripline plates to bedriven—thus directing, or kicking, the electron beam into two subsequent beamlines. This quad-driven stripline kicker is an eight-port electromagnetic network consisting of two actively drivenplates and two terminated plates. Electromagnetic measurements performed on the bi-kicker andquad-kicker were designed to determine: 1) the quality of the fabrication of the kicker, includingcomponent alignments; 2) quantification of the input feed transition regions from the input coax tothe driven kicker plates; 3) identification of properties of the kicker itself without involving the effectsof the electron beam; 4) coupling between a line current source and the plates of the kicker; and5) effects on the driven current to simulate an electron beam through the body of the kicker. Includedare the angular variations inside the kicker to examine modal distributions. The goal of the simulatedbeam was to allow curved path and changing radius studies to be performed electromagnetically. Thecold-test results were then incorporated into beam models.

Scott D. Nelson and James E. DunlapDefense Sciences Engineering DivisionElectronics Engineering

Figure 1. The quad-kicker in the Experimental Test Acceleratorbeamline as part of the verification experiments. Note the fourports on each end, which connect directly to the deflectionplates. Two of the white pulser cables are visible in the foreground.

Figure 2. The quad-kicker, tested using frequency- and time-domain scopes to cover the band for the swept frequency testsand instantaneous impulse tests.

470 Nelson_qk 7/27/99 8:00 AM Page 4-71


Progress

Kicker Port Testing

Each of the eight input ports of the kicker wastested over a frequency band from 45 to 500 MHz.Two ports connected to the input and output ofeach of the four plates through a tapered transition region, through a coaxial connector(Fig. 3). The pin on the plate connected directly tothe center pin of the coax. The results of themeasurements, shown in Fig. 4, indicate a broad-band match, with the exception of resonancescaused by the feed regions. The comparison inFig. 5 illustrates the feed region effects based onexperience from the bi-kicker and quad-kickerdevelopment activities. Figure 6 shows thecomplex input impedance of the kicker for one ofthe ports.

Cross-Coupling Terms

The cross-plate coupling terms of the kickercorresponded to the coupling between adjacentand opposite ends of the various plates to eachother. These terms represent energy that couplesfrom the kicker pulser-driven plate to those platesthat are terminated, thus inducing fields ontoplates that are not directly driven. These cross-coupling terms are appreciable (8% and 20%)even at the lower frequencies. Figure 7 shows themagnitude of the coupling between adjacent platesvs frequency. Figure 8 is a photograph of thequad-kicker plates.


Figure 3. Photograph showing four striplines of the kicker,each forming impedance-matched transmission lines. Thecenter pin that connects to the coaxial cable is visible at theend of each plate. The configuration at the other end is identical.

Inp

ut im

ped

ance

(Ω

)

Frequency (MHz)200

75

5025

100

125

150

300 50040010000

0.00

–2.00

–6.00

–8.00

–4.00

0 200 400MHz

0 200 400MHz

(a) (b)Figure 5. Kicker porttesting. The inputreflection coefficientvs frequency of thebi-kicker (a) wasmuch more uniformdue to the moregradual transitionregion after the coax-ial feeds. The quad-kicker (b) had a moreabrupt transitionafter the coax andhas a larger inputreflection coefficient.The spikes in the bi-kicker responsecurve are due to thegrounded plate reso-nances and wereeliminated in thequad-kicker.

Figure 4. Input impedance of each of the ports shown vsfrequency (margin of error ±0.3 Ω). The spikes at 388.75,414.0, and 460 MHz are higher-order mode resonances andcorrespond to Qs of 310 (quadrupole mode), 61 (dipolemode), and 29 (dipole mode), respectively.

470 Nelson_qk 7/27/99 8:00 AM Page 4-72


Forward-Coupling Terms

The kicker pulsers drive one end of the plates andthe other end is mated to reduce reflections on theplate structure. The loss along the plates is lessthan 1 dB. The transfer function from one end of theplate to the other is shown in Fig. 9.

Kicker Response

For identification of the transient properties ofthe kicker and the association between a simulatedbeam and the kicker ports, a ramp pulse

(0.95 V/300 ns) was used to excite the wire current.The resulting waveform that was induced on the downstream output port is shown in Fig. 10.

When the central wire representing an elec-tron beam was excited through the main body ofthe kicker, the “pump-up” time of the kicker wasobserved as an equivalent time constant of 70 ns. This corresponds to the cavity fill timebetween the simulated beam pulse and the ports.

Azimuthal Variations

During the course of the measurements, theazimuthal variation caused by the offset-rotationsof the current wire was measured and comparedagainst the theoretical solution for an offset wirein an ideal kicker. A comparison between thetheoretical solution (dashed line) for an electro-static coupling case and that for the experimen-tal cases at the peak coupling points of 68.4,139.2, and 209.4 MHz is shown in Fig. 11. Theangular frequency spectrum of the plots wastaken to determine the relationships between thevarious modes (dipole, quadrupole, and sextu-pole) in the kicker. The modal ratios are shown inTable 1.

The ratios are in a range similar to that determined by integrating the simple analytic representation5 along the plate boundaries shownin Eq. 1.

VV

qa

bd

qa

bd

a

bs

D= =

∫

∫

4 3

4

13

3

3

4

2

2

2

cos

cos

sinsin

;–

–

θ θ

θ θ

φφ

φ

φ

φ

φ

FY 98 4-73

0 0.2

0.5 1 2 5 10

Figure 6. Complex input impedance of the kicker for one of theports (the variance between the ports is ±7 Ω due to fabrica-tion differences). The three straight lines in the curve representunder-sampling.

0.50

0.40

0.20

0.10

0.30

0 250 500MHz

0.00

(a) (b)

5

1

7

3S

31

S71

S51

1V3

5

7

Figure 7. Magnitudeof coupling betweenadjacent plates vsfrequency, showingsignificant cross-coupling at multiplesof 80 MHz. The adjacent platecoupling is 20%; thecross plate couplingis 8%.

470 Nelson_qk 7/27/99 8:00 AM Page 4-73


(1)

where φ is the plate half-angle (39°), but 45° wasused, to be consistent with the theory, since Eq. 1assumes no gaps, b is the plate radius (12.87 cm),and a is the radius to the wire position (3.175 cm).Differences can be attributed to gap effects between

sinsin

V

Vab

Q

D= 1

22φφ

plates, end effects near the feeds, and simplifica-tions of the analytic representation.

Conclusions

Although the frequency range of interest forkicker applications is in the low hundreds ofmegahertz and is based on the bandwidth of thekicker pulser, there were initial concerns about


Figure 8. Identical quad-kicker plates connected to a 50-Ωcoaxial port. For experiments using the existing kicker pulsers,two of the plates were driven and two were terminated inmatched loads. Each plate is 78˚ wide (12.87 cm radius) andis supported by rexolite.

0

–10

–30

–40

–20

0 200 400MHz

Figure 9. Transfer function vs frequency from one end of theplates to the other. In the low frequency part of the spectrum,the curves for the various plates overlap to within 0.025 dB.

0.030

0.020

0.000

0.010

0 500 1000Time (ns)

0.20

0.15

0

0.05

0.10

0 18090 360270Degrees

Azi

mut

hal

var

iati

on

s

Figure 10. Effect of the 300-ns ramp pulse coupling from thewire current to one of the kicker plates monitored at a down-stream port. Notice that after about 70 ns, the coupling stabi-lizes to -0.005 V (-0.53%). The spikes occur at the transitionof each 300-ns excitation ramp waveform.

Figure 11. Comparison between theoretical coupling curve(dashed) based on electrostatic modeling of an ideal kickerstructure and experimental data.

Table 1. Angular frequency spectrum.

Static 68.4 139.2 209.4 Eq. 1

VQ /VD 0.136 0.134 0.140 0.169 0.174VS /VD 0.0252 0.0315 0.0327 0.0351 0.0203

470 Nelson_qk 7/27/99 8:01 AM Page 4-74


beam-induced effects. For this frequency range:the cross-coupling between adjacent ports is<14 dB; the input impedance for each port isbetween 50 and 90 Ω; transmission along the platesexperiences less than 1 dB of loss; cavity measure-ments show a cavity pump-up time, and a dI/dtcoupling between the current wire and the cavity.

The input reflection coefficient for some higherfrequencies can approach 30%; but these frequen-cies are expected to be outside of the normal operat-ing range of the kicker. However, in making themodifications from the bi-kicker design to thequad-kicker design, the frequency band where theseeffects make a pronounced difference was loweredand is closer to the operating band. Thus, subse-quent changes in the kicker design would need tobe made with this limit in mind. It should beemphasized however that the elimination of theshorted plates from the bi-kicker design substan-tially improved the operation of the quad-kicker.4

Acknowledgments

Thanks to B. Poole for numerous conversationsabout kicker development and to J. Chen and J. Weirfor their experimental activities.

References

1. Caporaso, G. J., Y. J. Chen, and B. R. Poole (1997),“Transmission Line Analysis of Beam Deflection in aBPM Stripline Kicker,” 1997 Particle AcceleratorConference, Vancouver, B.C., Canada, May, LawrenceLivermore National Laboratory, Livermore, California(UCRL-JC-126073).

2. Nelson, S. D. (1998), “Electromagnetic (Cold Test)Characterization of the Bi-Driven Kicker,”Lawrence Livermore National Laboratory,Livermore, California (UCRL-ID-129997), January, http://www-dsed.llnl.gov/documents/em/sdnkick98/.

3. Poole, B. R., G. J. Caporaso, and Y. J. Chen (1998),“Analysis and Modeling of a Stripline Beam Kickerand Septum,” 1998 Linear Induction AcceleratorConference, Chicago, Illinois, August 24-28.

4. Chen, Y. J., G. J. Caporaso, and J. Weir (1998),“Experimental Results of the Active Deflection ofa Beam from a Kicker System,” 1998 LinearInduction Accelerator Conference, Chicago,Illinois, August 24-28.

5. Chao, A. W. (1993), “Physics of Collective BeamInstabilities in High Energy Physics,” John Wiley and Sons, New York, New York, p. 6.

FY 98 4-75

470 Nelson_qk 7/27/99 8:01 AM Page 4-75

hotonic Doppler Velocimetry


Introduction

Laser Doppler velocimetry is a powerful diagnos-tic for the high-explosives physics community. Thefrequency of the Doppler-shifted light provides adirect measure of the instantaneous velocity of arapidly moving target illuminated by a laser. Lightfrom a multi-mode optical fiber illuminates thetarget and another multi-mode fiber collects theDoppler-shifted light, allowing spatially-resolvedvelocimetry information to be obtained.

Currently, physicists perform surface velocitymeasurements using a technique called Fabry-PerotVelocimetry. This system uses free-space Fabry-Perot interferometers and streak cameras for eachdata channel. These components are costly,complex, require maintenance and operator set-up,require a custom-built optical table and occupy aconsiderable volume. The Fabry-Perot system alsouses a large YAG laser whose output is doubled to532 nm because the streak camera photocathode isinsensitive in the infrared region of the spectrum.Although the Fabry-Perot velocimeter yields excel-lent data, overall channel count will always remainlow due to its size, cost, and complexity. As new NTShigh-explosives facilities become available to experi-menters, it will be highly desirable to have manyvelocimetry data channels available without thecost, complexity, and manpower-intensive set-uprequired for the current diagnostic system.

Our technique uses multi-mode fiber optics, anoptical PIN detector, RF electronics, and moderatesample-rate A/D converter technology. All of thecomponents fit into a small chassis. The advantage

of using multi-mode fiber is the significant increasein optical light collection from the target, comparedto that from a single-mode fiber. This advance inlaser Doppler velocimetry will enable the fielding ofsignificantly more data channels, greatly improvingthe spatial-temporal information obtained from thisdiagnostic at reduced cost, complexity, and experi-mental footprint.

A simplified version of this diagnostic will permitnanosecond-resolution shock arrival-time measure-ments by detecting the first incidence of Doppler-shifted light. Such a diagnostic will be very useful onhigh-fidelity flight tests in the weapons program atLawrence Livermore National Laboratory (LLNL).Future versions of this diagnostic could be madesufficiently compact and rugged to be practical forthe flight test application.

Figure 1 illustrates the basics of the photonicDoppler velocimetry system. A laser-generated opti-cal carrier propagates through a multi-mode fiber toa probe lens. The probe illuminates the target withthe optical carrier. As the target moves towards thelens, the reflected light is Doppler-shifted. The probelens collects a portion of the Doppler-shifted light,and the light propagates back through the multi-mode fiber. The Doppler-shifted light is mixed with afraction of the original optical carrier in a fiber-opticcoupler and is detected by a “square-law” opticaldetector. Under the appropriate polarization andmodal conditions, the square-law detector generatesan electrical current proportional to the square ofthe optical fields. For the Doppler-shifted light, thiscorresponds to a beat frequency proportional to theinstantaneous velocity of the target.

FY 98 4-77

We are developing a novel fiber-optic approach to laser Doppler velocimetry as a diagnostic forhigh-explosives tests. Using hardware that was originally developed for the telecommunicationsindustry, we are able to measure surface velocities ranging from centimeters per second to kilome-ters per second. Laboratory measurements and field trials have shown excellent agreement withother diagnostics.

Paul D. Sargis, Nicole E. Molau, and Daren SweiderDefense Sciences Engineering DivisionElectronics Engineering

475 Sargis_qk 7/27/99 8:06 AM Page 4-77


At high velocities, the beat frequency is too highto record directly on a transient digitizer. To over-come this limitation, we use a microwave phasediscriminator to measure the frequency-dependentphase shift induced by the Doppler signal. Recordingof the phase discriminator data is accomplishedusing a digitizer having a modest sampling rate.

Progress

In the past year, our efforts have focused on labo-ratory experiments and field trials.

The target for our first refereed test was a shock-driven copper foil, with the Fabry-Perot velocimeter

acting as the referee. As illustrated in Fig. 2, thecopper foil was in close proximity to a bridge wire,which was driven by a capacitive discharge unit(CDU). Green light from a frequency-doubled YAGlaser was focused onto the copper target through aprobe lens. The Doppler-shifted light was reflectedback through the probe lens and was simultaneouslyprocessed by both the Fabry-Perot velocimeter andthe photonic Doppler velocimeter.

The raw transient digitizer data from the photonicDoppler system is plotted in Fig. 3a, and the rawstreak camera data from the Fabry-Perot system isshown in Fig. 3b. The Doppler beat frequency signalin Fig. 3a was converted into frequency versus time,


Laser f

Coupler

v

Multi-mode fiberf

f + 2(v/c)f

2(v/c)f

f + 2(v/c)f

fProbelens

Opticaldetector

RFdetection

systemDigitizer

Figure 1. Basic blockdiagram of thephotonic Dopplervelocimetry system.

YAGlaser

Probe lens

PhotonicDoppler

velocimeter

Fabry-Perotvelocimeter

CDU

v

Copper foil

Bridge wirefo

fo

fo+∆f

∆f = 2*(v/c)*fo

Figure 2. Blockdiagram of thecopper foil velocityexperiment.

0.2

0 5Time (µs)

10 15

0.1

0.0

-0.1

-0.2

(a) (b)

V

Figure 3. Raw datafrom the copper foilexperiment: a) photonic Dopplervelocimeter; b) Fabry-Perotvelocimeter.

475 Sargis_qk 7/27/99 8:06 AM Page 4-78


and then into velocity versus time, as shown inFig. 4. Velocity values were hand-digitized from theFabry-Perot data and then plotted on the samegraph in Fig. 4. The negative velocity at the end ofthe data record is consistent with the rebounding ofthe copper foil after the shock event.

In a later experiment conducted at LLNL’s Site300, the copper foil target was replaced with analuminum plate that was driven by high explosives.The remainder of the experimental set-up was thesame as that illustrated in Fig. 2. This time,however, the Doppler beat frequency was too high tobe recorded directly on a transient digitizer. We used

FY 98 4-79

200

150

100

50

0

-502 4 6

Vel

oci

ty (

m/s

)

8 10 12 14Time (µs)

Photonic Doppler velocimeterFabry-Perot velocimeter

Figure 4. Processeddata from the copperfoil experiment.

1.2

1.1

1.0

0.9

Vel

oci

ty (

mm

/µs)

8642Time (µs)

Fabry PDV

Figure 5. Processed data from the Site 300 aluminum plateexperiment.

a microwave phase discriminator to measure thefrequency-dependent phase shift of the incomingsignal. The processed result is plotted with theFabry-Perot result in Fig. 5.

We also conducted a series of measurements inthe laboratory using a continuously moving target,which consisted of a speaker driven by an audiooscillator. This set-up provided the framework forevaluating system stability and reliability. Wediscovered that the signal-to-noise ratio varieswith time and is adversely affected by physicallymoving the multi-mode fiber. The likely causes ofthis phenomenon include: instability of the opticalstate on the polarization optical detector,mode selective loss, and de-phasing of the opticalsignals caused by the surface reflection and opticalfiber propagation.

Future Work

We plan to conduct further research into systemstability and reliability issues, including makingdetailed measurements of the changes induced inthe optical polarization state, the mode populations,and their relative phases by the moving surface andthe optical elements in the system. Once we havegained a better understanding of these issues, wewill design and build an optimized system. Finally,we will demonstrate the optimized system on anAsay foil experiment.

475 Sargis_qk 7/27/99 8:06 AM Page 4-79

odeling Coupled Heat and Mass Diffusion


Introduction

Drying is the removal of water from a solid byevaporation. In the evaporation process, heat has tobe supplied to the material, so that simultaneoustransfer of heat and mass occurs. Several dryingperiods are defined:

1. First Drying Period (constant rate drying). Thesurface of the solid is covered with a continu-ous layer of free water, and evaporation takesplace mainly at the surface. The rate of dryingdepends entirely on parameters external to thesolid, such as the air velocity, temperature,and humidity. If the external conditions areconstant, then the drying rate is constant. Thetemperature of the solid will equilibrate to thewet bulb temperature of the air.

2. First Falling Rate Period. As drying proceeds,the fraction of wet area decreases withdecreasing surface moisture content. Thesurface will form discontinuous wet patchesand the mass transfer from the surface willdecrease with a slowly rising surface tempera-ture. Free water still exists at the surface, the“dry” patches still contain bound water, andthe vapor pressure at the surface is deter-mined by the Clausius-Clapeyron equation.

3. Second Falling Rate Period. No free waterexists at the surface. The surface temperaturewill rise rapidly, during which a receding evap-oration front appears, dividing the solid into awet region and the sorption (bound water)region. Inside the evaporation front, the mater-ial is wet (that is, the voids contain freewater), and the main mechanism of moisturetransfer is capillary flow. Outside the front, allwater is in the sorptive or bound state and themain mechanisms of moisture transfer arebound water and vapor diffusion. Evaporationtakes place at the front as well as in the wholesorption region, while vapor flows through thesorption region to the surface.

In our application all the water is in the“sorptive” or “bound” state, and the main mech-anisms of moisture transfer are bound water andvapor diffusion.

Progress

The most appropriate and applicable model formass transfer that seems to describe well the rele-vant physics over the entire range of pressure—fromatmospheric pressure where continuum dynamicsprevails, on down to hard vacuum conditions at

FY 98 4-81

We have developed a model that includes thermally-driven mass diffusion to predict the evapora-tion rate of bound water in a porous material in a vacuum. All water in the porous material is in the“sorptive” or “bound” state and the main mechanisms of moisture transfer are bound water andvapor diffusion. This differs from the state in which free water exists in the pores and moisture trans-fer is by capillary flow.

Arthur B. ShapiroNew Technologies Engineering DivisionMechanical Engineering

Philip M. GreshoAtmospheric Science DivisionEnvironmental Programs

480 Shapiro_qk 7/27/99 8:09 AM Page 4-81


which free-molecule flow prevails—is a model thatwas first envisaged by James Clerk Maxwell about140 years ago, during the early development of thekinetic theory of gases. It was later rediscoveredseveral times by others. Called (in recent times) thedusty gas model1 (DGM), it treats the porousmedium (the “dust”) as a collection of fixed spheresaround which “n” ideal gases flow. The DGM is mostsuccinctly described for isothermal conditions asfollows:

(1)

where Ni is the molar flux (ciui) of component i(g-moles/cm2 s), xj is the mole fraction of componentj, P is pressure (dyne/cm2), T is temperature (K), R isthe gas constant (~8.3 x 107 gm cm2/g-mole s2 K),and the diffusivities are as follows:

1. is the effective Knudsen (free-molecule) diffusivity, Ko is an effective diffu-sion length (Ko = 2a/3 for a tube of radius a),in cm, and

(2)

is the mean molecular speed (Mi is the molec-ular weight);

2. is the effective ordinary (bulk)binary diffusivity associated with continuum(Stefan-Maxwell) diffusion, where K1 is adimensionless constant (K1 ≤ 1) and Dij is the(known) binary molecular diffusivity;

3. Dv = BoP/µ is the viscous/Darcy diffusivity,where µ is the gas viscosity (gm/cm s) and Bois the effective “area” of the medium (cm2; Bo= a2/8 for a tube of radius a—Poiseuille flow).

For a porous medium, the DGM is a three-para-meter model, in which the three parameters that“characterize” the porous medium (Ko, K1, Bo) arebest determined experimentally. Note that the DGMsuccessfully combines three physical phenomena inone model: 1) free-molecule flow (Knudsenflow/diffusion) that prevails at very low pressure(where Dij → ∞ and Dv → 0); 2) ordinary (Stefan-Maxwell) diffusion that is important at higher pres-sures ( ); and 3) viscous flow, whichcarries all species at the same rate (non-separative)and is most simply “described” by the first term inthe left-hand side and the last term in the right-hand

D Deije

i / → ∞

D K Dije

ij = 1

υ πi iRT M≡ 8 /

D Kie

o i = υ

= − ∇( ) − ∇1RT

x Px D

RTDPi

i v

ie

N

D

X N X N

D

i

ie

j i i j

ije

j i

n

+≠

∑ –

side of Eq. 1, via , whichbasically describes Poiseuille flow.

In addition to the transport of water in the porousmedia defined by Eq. 1, an expression is needed forthe evaporation rate of liquid water to gas in avacuum.2 Kinetic theory applies to systems invacuum. Equilibrium requires that the rate of evapo-ration equals the rate of condensation. Therefore,we can determine the evaporation rate by equatingit to the mass of molecules striking the surface.

From kinetic theory, the number of molecules of agas that strike a surface is

(molecules/cm2 s),

where the average gas velocity is given by Eq. 2, andthe number of molecules per volume is determinedby the ideal gas law

(molecules/cm3),

where NA is Avogadro’s number (6.0228 × 1023

molecules/gmole), R is the universal gas constant(6.236 × 104 cm3 mm/g gmole K), Pmm is the watervapor pressure in mm of Hg, and T is the tempera-ture in K.

The mass of gas striking (or leaving) a surfaceper unit area per unit time is then

.

Using the molecular weight of water as M = 18.02g/gmole, the expression for the mass of water vaporleaving a surface, G, can then be evaluated to be

(g/cm2 s).

Over a small temperature range, the vapor pressureof water may be represented by the Clausius-Clapeyron equation

,

where A and B are constants determined fromexperimental data such as shown in Table 1.

Using the end points in Table 1, A and B are deter-mined to be A = 17.765 and B = 5314.8. Then, thefinal expression to calculate the evaporation rate is

P ABTmm exp – =

G

P

Tmm .

/= 0 247

1 2

=

=

/ /MN

N PRT

RTM

P MRTA

A mm mm14

82

1 2 21 2

π π

G mv m n i= =

14

υ

nN P

RTA mm =

ν υ= 1

4n a

N x D P RTiv

i v – /≅ ∇




in terms of the surface temperature is

(3)

Equation 3 applies to a completely wet liquidsurface. This expression must be multiplied by thefraction of surface area that is wet to give the evapo-ration rate for a partially wetted surface.

Future Work

Experiments are planned that will determine thethree parameters of the DGM for one or more partic-ular cases of current interest. Also, the restriction toisothermal flow will be removed. This is most easilyintroduced at the low pressure end (vacuum

G

T

T=

−

6 27

17 7655314 8

1 22.

exp ..

( )./

g/cm s

conditions, where it is most important) by addingthe following term to the right-hand side of Eq. 1:

,

which accounts for thermal transpiration (bulk flowowing to temperature gradients) and, interestingly, isperhaps somewhat counter-intuitive in that thermally-induced flow is up the temperature gradient.

Our ultimate goal is to include these develop-ments in the TOPAZ heat transfer code to extend thecode’s capability into the arena of low-speed flowand heat transfer of ideal gaseous mixtures inporous media, especially at low pressure.

References

1. Mason, E., and A. Malinauskas (1983), Gas Transportin Porous Media: The Dusty Gas Model, Elsevier,Amsterdam.

2. Dushman, S. (1966), Of Scientific Foundations ofVacuum Techniques, 4th ed., John Wiley and Sons,New York, New York.

3. Hougen, O. (1959), Chemical Process Principles,2nd ed., John Wiley and Sons, New York, New York,p. 82.

+ ∇x P

RTTi

2 2

FY 98 4-83

Table 1. Pressure of aqueous vapor over water.3

T P(K) (in. of Hg)

288.6 0.5218294.1 0.7392299.7 1.0321


nalysis and Modeling of a Stripline Beam Kicker and Septum


Introduction

The stripline kicker is designed to spatially sepa-rate a high-current electron beam for transport intotwo separate beamlines. However, to provide asignificant angular kick to the beam, a magneticdipole septum is required. This system is shownschematically in Fig. 1.

The operation of the system is as follows: a high-voltage pulse is applied to the downstream ports ofthe kicker, and the beam is spatially separated(kicked) by a combination of the transverse electricand magnetic dipole forces associated with thetransverse electromagnetic (TEM) waves propagat-ing on the strip transmission lines.

The beam is then directed into a septum magnetwith opposite polarity dipole fields on either side ofthe plane separating the two downstream beam-lines. All the upstream ports and the two down-stream ports in the non-kick plane are terminated ina matched load impedance for the dipole transmis-sion mode on the structure. It should be noted thatsteering in both planes can be accomplished by alsodriving the other pair of plates.

Kicker TEM Fields and Beam Deflection

To steer the beam in x, opposite polarity high-voltage pulses are applied to the downstream portsin the y = 0 plane. The potential within the kickerplates (r < b) is given by

, (1)

where b is the interior radius of the kicker plates,and φ0 is the angle subtended by the kicker plates.

The voltage applied to the plate is Vp, giving atotal steering voltage of 2Vp. The solution is deter-mined by solving for the potential in the region r < b,and using the boundary conditions that 1) the poten-tial at r = b is given by the appropriate applied platevoltages, and 2) the potential in the gaps betweenthe plates is zero. The TEM fields can be easilyderived from this scalar potential.

VV

m

mm

r

b

p

m odd

m

sin cos

=

×

( )

=∑

4

1

20

πφ φ

FY 98 4-85

A fast stripline beam kicker and septum are used to dynamically switch a high-current electronbeam between two beamlines. The transport of the beam through these structures is determined bythe quality of the applied electromagnetic fields as well as temporal effects due to the wakefieldsproduced by the beam. In addition, nonlinear forces in the structure will lead to emittance growth.The effect of these issues is investigated analytically and by using particle transport codes. Due to thedistributed nature of the beam-induced effects, multiple macro-particles (slices) are used in the parti-cle transport code, where each slice consists of an ensemble of particles with an initial distribution inphase space. Changes in the multipole moments of an individual slice establish electromagneticwakes in the structure, and are allowed to interact with subsequent beam macro-particles to deter-mine the variation of the steering, focusing, and emittance growth during the beam pulse.

Brian R. Poole, Lisa Wang, and Yu Ju (Judy) ChenDefense Sciences Engineering DivisionElectronics Engineering

George J. Caporaso Laser Programs

TA 485 Poole 7/27/99 8:12 AM Page 4-85


The m = 1 term in Eq. 1 represents a transversedipole force which provides the beam steering, whilethe higher order terms will contribute to emittancegrowth in the beam. The beam deflection due to thecombined electric and magnetic dipole forces isgiven by

, (2)

where the critical current, Ic, is defined by

(3),

and I0 = 17 kA, Z0 = 377 Ω, L is the length of thekicker, b is the inner radius of the kicker plates, Zkis the kicker impedance, and γ is the usual relativis-tic factor.

Progress

Dipole Wake Impedance and Beam-Induced Steering

In our application, the beam current is suffi-ciently large to induce substantial voltages andcurrents on the strip transmission lines. These volt-ages and currents are introduced on the transmis-sion lines as the beam traverses the upstream anddownstream gaps, and from changes in the dipolereturn current as the beam is deflected. A detailedmodel has been described previously.1,2

The m = 1 transverse dipole wake impedance forthis structure3 is given by

. (4) ×

+

sin sin cos 2 ω ω ωLc

jL

cL

c

ZcZ

bk

⊥ ( ) =

ωπ

φω

sin 8

21

2 22 0

II Z

ZbLc

k

sin

=

π γ βφ16

2

20

2 0

02

∆xbV

I Zp

c k

sin= ( )

πφ0 2

The imaginary part of the dipole impedance, Ζ⊥0 = Im[Z⊥ (ω = 0)], is a measure of the asymptoticbeam deflection due to the beam-induced fields. Ithas been shown that the asymptotic beam deflectionfor an initially offset beam with current IB injectedinto the kicker has the form1

, (5)

where x0 is the injection offset of the beam. It iseasily shown for sufficiently small beam currentsthat

. (6)

To examine the relevant physics issues, a kickerhas been designed and installed on LawrenceLivermore National Laboratory’s Experimental TestAccelerator (ETA-II). The ETA-II kicker has thefollowing set of parameters: b = 12.87 cm, φ0 = 78°,Zk = 50 Ω, L = 164 cm, and Ic = 3.9 kA. The outerenclosure has a radius of 19 cm. To verify the valid-ity of the transmission line model, the structurewas modeled using LLNL’s TIGER 3-D time-domainelectromagnetic code to determine the dipoleimpedance spectrum.

Figure 2 shows a comparison of the dipoleimpedance as calculated from Eq. 4 with numericalresults from the TIGER code for the ETA-II kicker. Ascan be seen, there is good agreement between thetransmission line model and the 3-D code results.The differences can be attributed to end cavity effectsassociated with the feeds to the external ports andeffects due to higher order modes in the structure.

The effect of the wake impedance on the deflec-tion of the beam can be quite dramatic. For exam-ple, for a 6-MeV, 2-kA beam initially offset by 2 cmgoing into the kicker, the tail of the emerging beamwill be offset by 3.1 cm at the exit of the structure.These effects have been observed experimentally4

and are consistent with theory.

x xL

I ZI ZB∞ ⊥≈ +

0 20 0

012π

γ β

x xII

B

c∞ =

cosh 0

2


Termination

Termination

– Drive pulse

+ Drive pulse

– Driven electrode

+ Driven electrode

z

x

Beam

Figure 1. Kicker andseptum configura-tion for dynamicbeam-steering inone plane.

TA 485 Poole 7/27/99 8:12 AM Page 4-86


Nonlinear Forces and Emittance Growth

From Eq. 1 it is seen that for a dipole excitationof the kicker, all higher order odd multipoles will beexcited in the structure. Although the higher ordermultipoles reduce in strength as (r/b)m, it is possi-ble under certain conditions that the beam willexperience these fields, especially the m = 3 sextu-pole component.

For example, with ETA-II parameters the space-charge fields due to the beam may require that thebeam entering the kicker have a large radius toenable the downstream beam to be at or near a waistwhen entering the dipole septum. This is important tominimize any emittance growth due to the nonlinearfields associated with the septum magnet.

To estimate the effect of the higher order multi-poles due to the kicker on the beam emittance, asimple particle transport code was developed. Thecode includes the external fields in the kicker regionas defined by Eq. 1, and is being expanded toinclude the beam-induced effects and space-chargeeffects self-consistently.

Presently, the particles respond only to the exter-nal fields. However, we can estimate the emittancegrowth in the structure by using the external fieldsonly. As an example, using the ETA-II kicker previ-ously described, a 6-MeV beam is injected into thekicker with an unnormalized edge emittance of13 cm-mrad and a convergence angle of 0.03 rad.The injected beam radius is 4 cm, which allows thebeam to experience the higher order multipoles.Figure 3 shows a configuration-space image of theemerging beam from the kicker showing a centroidlocation of 2.7 cm, consistent with Eq. 2. The trian-gular image has been observed experimentally.4

Despite the strong deformation of the beamimage, the emittance growth is predicted to be about53% for this beam. However, transport calculationshave also shown that it is possible to transport asmaller radius high-current beam, ~2 kA, throughthe kicker giving an estimated emittance growth onthe order of 2%.

Magnetic Dipole Septum

The septum magnet provides an additionalangular kick to the beam as it emerges from thekicker. The kick is in opposite directions on eitherside of x = 0. The septum is shown schematicallyin Fig. 4.

The dipole magnetic field required to produce abeam exit angle of ~15° is determined from

, (7)

where is the axial length of the magnet, θi is theincident beam angle, θi + ∆θ is the desired exitbeam angle, and Eb is the beam energy.

A preliminary design for a dipole septum magnetto be used with the ETA-II kicker is being developed.The parameters for the design are an axial length of20 cm, and a magnetic field of about 276 G for abeam energy of 6.3 MeV. The dimensions of theaperture are about h=6 cm high and w=31 cm wide.Careful optimization of the design is required tominimize possible emittance growth in the transitionregion where the field changes sign.

l

× +( ) ( )[ ] sin – sinθ θ θi i∆

B

mce

E

mc

E

mcb bl =

+

2

2

22

FY 98 4-87

-1000

1000

2000

0Imag

(Z⊥

1)(Ω

/m)

Frequency (MHz)

2000 600400


Figure 2. Dipole wake impedance.

-2

-1

0

1

2

y (c

m)

x (cm)210 6543

Figure 3. Configuration-space image of beam emerging fromkicker experiencing a large sextupole field.

TA 485 Poole 7/27/99 8:12 AM Page 4-87


Magnetic modules are being developed for parti-cle transport codes to estimate the emittancegrowth through the septum magnet. With carefuldesign, including shims, preliminary estimates showan emittance growth on the order of 4% through theseptum magnet.

Conclusions

Self-consistent models are being developed formodeling the transport of high-current, space-charge-dominated beams through fast beam kickersand dipole septum magnets. The effect of beam-induced forces due to the wakefields of the beam areincluded in the analysis.

In addition, emittance growth due to nonlinearforces associated with higher order multipoles inboth the kicker and septum have been estimated.Preliminary estimates of beam-induced steering areconsistent with the experimental program.4 Theeffect of space charge, image forces, and fringefields in the structures have yet to be included.

Acknowledgments

Discussions on the electromagnetic characteriza-tion of the kicker with S. Nelson are greatly appreci-ated. Thanks also go to D. Steich for running the 3-Delectromagnetic simulations of the kicker.

References

1. Caporaso, G. J., Y. J. Chen, and B. R. Poole (1997),“Transmission Line Analysis of Beam Deflection in aBPM Stripline Kicker,” 1997 Particle AcceleratorConference, Vancouver, B.C., Canada, May 12–16,Lawrence Livermore National Laboratory, California(UCRL-JC-126073).

2. Poole, B. R., G. J. Caporaso, and W. C. Ng (1997),“Wake Properties of a Stripline Beam Kicker,” 1997Particle Accelerator Conference, Vancouver, B.C.,Canada, May 12–16, Lawrence Livermore NationalLaboratory, California (UCRL-JC-126075).

3. Ng, K.-Y. (1988), “Impedances of Stripl ine Beam-Position Monitors,” Particle Accelerators23, pp. 93–102.

4. Chen, Y. J., G. J. Caporaso, and J. Weir (1998),“Experimental Results on the Active Deflection ofa Beam from a Kicker System,” XIX InternationalLinac Conference (Linac98), Chicago, Illinois,August 23–28.


W

Beam separation

hB B

Figure 4. Magnetic dipole septum magnet.

TA 485 Poole 7/27/99 8:12 AM Page 4-88

Harry E. Martz, Center Leader

Center for Nondestructive Characterization

The Nondestructive Characterization (NDC)Center (formerly the Nondestructive EvaluationThrust Area) at Lawrence Livermore NationalLaboratory (LLNL) supports initiatives that advanceinspection science and technology. Our goal is toprovide cutting-edge technologies that show promisefor quantitative inspection, and characterizationtools three to five years into the future.

The NDC Center strategic objectives involvequantitative NDC, and fast nanometer-scale imaging.This past year the NDC Center portfolio of projectsfocused on quantitative NDC in support of weaponssystem performance and laser systems.

The NDC Center supports a multidisciplinaryteam, consisting of mechanical and electrical engi-neers, physicists, material and computer scientists,chemists, technicians, and radiographers. Theseteam members include personnel that cross depart-ments within LLNL. Some are from academia andindustry, within the United States and abroad. Thiscollaboration brings together the necessary anddiverse disciplines to provide the key scientific andtechnological advancements required to meet LLNLprogrammatic and industrial NDC challenges.

NDC provides materials characterization inspec-tions of finished parts and complex objects, to find

flaws and fabrication defects and to determine theirphysical and chemical characteristics. In addition,applying NDC throughout the life cycle of a partsaves time and money and improves quality. Forexample, NDC is being applied at the beginning of apart to develop new materials and to aid in processdesign and development. NDC encompasses processmonitoring and control sensors and the monitoringof in-service damage. NDC is also being applied atthe end of a part’s life for proper reuse or safe andproper disposal. Therefore, NDC is becoming both afront-line and an end-of-line technology that stronglyimpacts issues of certification, life prediction, lifeextension, reuse, and disposal.

To meet today’s programmatic demands, it isimportant to increase collaboration among LLNLengineering centers, departments, and programs.This year, collaborators included theMicrotechnology Center, Defense Systems, MaterialsScience and Technology, Physics, the Life ExtensionProgram, the National Ignition Facility Project,Computations, and Nonproliferation. Such collabora-tions enable us to stay at the leading edge of NDCtechnology, research, development, and applicationin support of LLNL programs.


Center for NondestructiveCharacterization

5


Contents

5. Center for Nondestructive Characterization

OverviewHarry E. Martz, Center Leader

Techniques for Enhancing Laser Ultrasonic Nondestructive EvaluationGraham H. Thomas, Robert D. Huber, Diane J. Chinn, James V. Candy, and James Spicer ........................5-1

Nondestructive Evaluation of an Aluminum Alloy Using Hyperspectral Infrared Imaging Randy S. Roberts .......................................................................................................................................5-9

In-Situ Identification of Anti-Personnel Mines Using Acoustic Resonant SpectroscopyRandy S. Roberts and Roger L. Perry.......................................................................................................5-13

An Acoustic Technique for the Non-Invasive In-Situ Measurement of Crystal Size and SolutionConcentrationDiane J. Chinn, Paul R. Souza, and Harry F. Robey..................................................................................5-19

Micro X-Ray Computed Tomography for PBX CharacterizationDiane J. Chinn, Jerry J. Haskins, Clinton M. Logan, Dave L. Haupt, Scott E. Groves, John Kinney, and Amy Waters..................................................................................................................5-23

Evaluation of an Amorphous Selenium Array for Industrial X-Ray ImagingClinton M. Logan, Jerry J. Haskins, Kenneth E. Morales, Earl O. Updike, James M. Fugina, Anthony D. Lavietes, Daniel J. Schneberk, Gregory J. Schmid, Keo Springer, Peter Soltani, and Kenneth Swartz.................................................................................................................................5-27

LANDMARC Radar Mine DetectionStephen G. Azevedo, Jeffery E. Mast. James M. Brase, and E. Tom Rosenbury ........................................5-39

IMAN-3D: A Software Tool-Kit for 3-D Image AnalysisSailes K. Sengupta...................................................................................................................................5-51

Image Recovery Techniques for X-Ray Computed Tomography in Limited-Data EnvironmentsDennis M. Goodman, Jessie A. Jackson, Maurice B. Aufderheide, and Erik M. Johansson.......................5-61


echniques for Enhancing Laser UltrasonicNondestructive Evaluation


Introduction

Although the use of ultrasound for nondestructiveevaluation is a mature technology, there continue tobe many advances that expand its role in materialcharacterization, manufacturing process control,defect detection, and life cycle management.Ultrasonics is evolving with improvements such ashigher frequencies to sense smaller defects, modernsignal processing methods to increase sensitivity,classification algorithms for defect characteriza-tion and the most recent imaging techniques todisplay defects.

Despite these advances, a universal limitationof traditional piezoelectrically generated anddetected ultrasound is the need to transmit theacoustic energy from the transducer into the partthrough a fluid, most often water. For many partsand materials, particularly those of interest toLLNL and the Department of Energy (DOE), it isextremely desirable to eliminate this couplant.

Laser generation and detection of ultrasonicenergy provides a method to perform remote, non-contact ultrasonics.1 It allows ultrasonic evalua-tions in high-temperature and radioactive environ-ments, in applications where access is restricted,such as in a vacuum, and on materials that wouldbe damaged by couplant contamination. For ultra-sonic inspections on radioactive materials, anycouplant used becomes hazardous waste, and thuslaser ultrasonics reduces hazardous waste since nocouplant is required.

Applications for laser-based ultrasoundcontinue to be implemented as breakthroughs inthe technology occur, but there is still much to beunderstood before its full potential can be realized.A significant limitation of laser-based ultrasound isits poor sensitivity as compared to the sensitivity oftraditional piezoelectric-based ultrasonics. Toimprove the sensitivity, research is being pursuedin the areas of improved ultrasonic generation,better detectors, and signal processing to make

FY 98 5-1

Ultrasonic nondestructive evaluation has been an extremely powerful tool for characterizingmaterials and detecting defects for many years. Piezoelectric transducers, traditionally used togenerate and detect high-frequency acoustic energy, usually require a liquid medium to couple theultrasound into the material being characterized. This need for a couplant restricts the applicabilityof ultrasonics since many materials can be damaged by the use of couplants. We are developing atechnology that generates and detects the ultrasonic pulses with lasers and thus there is norequirement for couplants. The ultrasound is generated and detected in a remote, non-contactmanner since only the laser light is in contact with the material. Laser-based ultrasound has wideapplication in many Lawrence Livermore National Laboratory (LLNL) programs, especially whenremote and/or non-contact sensing is necessary.

Graham H. Thomas, Robert D. Huber, and Diane J. ChinnManufacturing and Materials Engineering DivisionMechanical Engineering

James V. CandyElectronics Engineering Technologies DivisionElectronics Engineering

Professor James SpicerJohns Hopkins UniversityBaltimore, Maryland

505 Thomas_qk 7/27/99 8:51 AM Page 5-1


laser ultrasonics viable. Also beam-forming andother signal processing techniques have beendeveloped to improve the defect detection levels oflaser acoustics. Significant improvements in boththe generation and detection aspects of LLNL’slaser-based ultrasound capability were realizedduring this project. Laser-based ultrasound willcontinue to have an increased role in nondestruc-tive evaluation as the sensitivity limitation is solvedthrough research.

Progress

This project is exploring the science of generat-ing acoustic energy with a laser pulse, and themethodology of detecting ultrasonic signals withlaser interferometers. The goal is not to replaceexisting piezoelectric-based ultrasonics with lasergeneration and detection, but to supplement thetechnology to expand its role in LLNL programs.The following text describes the equipment andsoftware algorithms for laser-based ultrasonicsdeveloped by this project.

Signal and Image Processing and Beam-Forming

A large portion of our effort has been in the areaof signal and image processing, an approach toimprove the sensitivity of laser-based ultrasonics. Ithelps extract more information from the experimen-tally obtained data. In the past we have demon-strated the benefits of several signal processingapproaches that greatly improve laser-based ultra-sonic sensitivity to finding defects.

A model-based signal processing technique hasbeen developed and tested.2,3 This techniquepredicts the acoustic signals that are generated by aspecific laser. A code (WAVER) developed at theJohns Hopkins University4 has been modified tohandle materials and laser configurations that are ofinterest to LLNL programs. The model-based signalprocessing was implemented on ultrasonic signalsgenerated with a Nd:YAG laser and detected with aMichelson interferometer.5 This demonstrated tech-nology is significant to the implementation of laser-based ultrasound for many nondestructive evalua-tion applications since it allows the application oflaser-based ultrasonics for materials where acousticattenuation is large.

Another signal processing approach that we havedemonstrated is beam-forming. Beam-formingimproves defect detection sensitivity by viewing the

defect from several directions and combining theinformation from each direction in a manner toaccentuate the defect’s image. Previous workmodeled the acoustic beam from an array of sendersand receivers. This year the predicted defect sensi-tivity was confirmed with beam-forming algorithmsprocessing experimental data. This methodenhances the detection and display of defects bycombining the information from an array ofsensors.5,6 We have modeled the array and itsresponse to a flaw.

Based on these results, we configured a syntheticlaser ultrasonic array and confirmed the model. Thisnew technology is one of our methods for processingultrasonic data.

This past year we have developed matched-fieldimaging of the laser-based ultrasound signals. Thissignal processing technique uses a novel correlation-canceling approach to eliminate noise, therebyincreasing the signal-to-noise ratio of the experimen-tally obtained data. Eleven linear scans of Nd:YAGpulsed-laser-generated, laser-based Michelson-interferometer-detected data were obtained on a9.5-mm-thick aluminum plate. Each scan consists of21 amplitude vs time waveforms.

Figure 1 shows the source and detection loca-tions for these scans. There were 11 source loca-tions spaced 2 mm apart, and 21 detection locationsspaced 1 mm apart. The source position was heldfixed for a scan (21 different detection locations),and then moved 2 mm for the next scan. A 1/16-in.(1.6 mm)-diameter hole was then drilled into thealuminum plate to simulate a defect, and the platewas scanned over the same region that was scannedprior to the hole being drilled.


Detection sites1-mm spacingHole

Source sites2-mm spacing

Figure 1. Source and detection sites for laser ultrasonic genera-tion and detection on an aluminum plate with hole. Dataobtained before and after the hole was drilled were used toobtain data in Figure 2.

505 Thomas_qk 7/27/99 8:51 AM Page 5-2


The waveforms obtained after the hole wasdrilled, shown in Fig. 2a, were used along with thereference waveforms obtained before the hole wasdrilled (Fig. 2b). An optimal correlation-cancelingscheme was developed to extract just the hole infor-mation, as shown in Fig. 2c, with a single channelresult shown in Fig. 2d.

Processing yielded the canceled signal from thepre-hole and post-hole data. These signals wereused to generate images, which show the effects ofthe presence of the hole on the ultrasound for differ-ent source and detector locations (Fig. 3). Thisdemonstrates how laser-based ultrasound withsignal processing can be used for defect detectionand location.

Facility

Significant improvements in our laboratory havebeen realized over the duration of the project. At thestart, LLNL had a Michelson interferometer and asmall Nd:YAG laser for laser-based ultrasound work.These items had only limited application, and therewas only a small area devoted to laser-based ultra-sound. Once the project began, improvements to thecorrection circuit of the Michelson interferometerwere made that greatly increased its stability.Modifications to both the electronic and mechanicalcomponents were made.

We have added a LISOR (Light In, Signal OutReceiver) interferometer to expand our detection

FY 98 5-3

0 1

Time (µs)

Reference

Measurement

Cancelled

0 2 4 6 8Time (µs)

Cancelled(d)

(b)

(c)

(a)

ReferenceMeasurement

Figure 2. Laser ultrasonic correlation cancelation for enhanced flaw detection. Top left plot (a) shows the waveforms obtained afterthe hole was drilled (Measurement) for one source location. Top right signals (b) were obtained prior to drilling of the hole(Reference). Bottom left plot (c) displays the cancelled signals. Bottom right display (d) shows an individual wave for the pre- andpost-drilled cases with the corresponding cancelled signal.

505 Thomas_qk 7/27/99 8:51 AM Page 5-3


capabilities. This instrument is a Fabry-Perot-basedsystem that complements the earlier path-stabilizedMichelson interferometer. The Michelson requireshighly reflective surfaces to sense the ultrasonicsignal, whereas the Fabry-Perot works on lessreflective surfaces.

This system includes an interferometer and alaser. The laser is a frequency doubled Nd:YAG,which has an output power of 200 mW at a wave-length of 532 nm. Fabry-Perot interferometers canuse light scattered from rough surfaces since theycan work with multiple speckles. Michelson interfer-ometers work with a single speckle only, whichlimits their use to polished surfaces.

The LISOR system has been used successfully inthe laboratory to detect ultrasound propagatingthrough various materials and specimens. To

increase the flexibility of the ultrasonic detectionsystem, an optical fiber cable 20 m long transmitsthe laser light from the interferometer’s laser to thespecimen, allowing the testing of items up to thatdistance away from the laser and interferometer.

Fiber optics have the advantages of allowing largedistances between equipment and test location,providing access to hostile or hard to reach environ-ments, changing test configuration easily, and elimi-nating the need for line of sight between laser andpart. The Michelson interferometer has a frequencyrange of DC to 40 MHz, while the Fabry-Perot has afrequency range of 3 to 100 MHz. Together, theseinterferometers provide a full range of detectioncapabilities for most ultrasonic evaluation needs.

Another addition to the laser ultrasonic facility isa much improved source laser that provides more


Laser source

Receiving

FLA

FLA

FLA

FLA

Figure 3. Images obtainedfrom correlationcancellation. The 11images correspond to11 different sourcelocations for linearscans on thealuminum plate. The hole appearssharpest when thesource is directly overthe hole.

505 Thomas_qk 7/27/99 8:52 AM Page 5-4


light energy, a higher repetition rate and a uniformspot. This Nd:YAG source laser can operate at twowavelengths and has much better beam quality thanthe smaller Nd:YAG laser. This laser has made atremendous difference in the signals obtained, and isaiding greatly in the model-based processing workbeing performed at LLNL.

As part of the data acquisition and imagingeffort, we acquired computer-controlled motionstages. These stages include two translation stagesand one rotation stage that allow the movement ofspecimens for scanning. Data acquisition andmotion control programs have been written tocontrol the stages and capture the ultrasonicsignals. The signal processing and beam-formingalgorithms are combined with this software torender images of defects from the data obtainedwhen specimens are scanned. Most nondestructiveevaluation tasks require the results to be an imageof the defects. Scanning capability is needed togenerate ultrasonic images.

Fiber Optics

Fiber optics are a significant development forlaser-based ultrasonic implementation since theyallow lasers and other equipment to be located inone place, with the object under test located inanother place. The laser light can then be transmit-ted from one location to the other, completelycontained in the fibers, thus eliminating the hazardsof transmitting beams through the air. Fiber opticshave been developed and implemented in severalapplications. The first such activity involved placingthe probe arm of the Michelson interferometer in asingle mode optical fiber.

Figure 4 is a comparison of signals obtainedusing the Michelson interferometer for the twocases of 1) optical fiber probe path, and 2) no opti-cal fiber. The two waveforms show the ultrasounddetected in an aluminum plate for a through trans-mission case for a laser-generated thermoelasticsource. The thermoelastic source causes no damageto the sample and is therefore useful for nondestruc-tive evaluation and characterization.

The waveforms are quite similar, with only aslight decrease in the signal-to-noise for the case ofthe fiber, because there is some loss of light associ-ated with the use of the optical fiber. Although thereis a slight decrease in the signal-to-noise ratio forthe case of optical fiber in the interferometer, theremay be situations where the benefits of fiberoutweigh the decrease in signal-to-noise.

Fiber optics are also used with the Fabry-Perotinterferometer. Initially the illuminating laser wastransmitted through air and only the light reflectingoff the specimen was transferred to the Fabry-Perotinterferometer in a multi-mode fiber. Then wedemonstrated that the light from the part-illuminatingsource laser could be transmitted through a multi-mode fiber. Thus all of the travel path of the laserlight can be contained in optical fiber.

Fiber optic transmission of the laser beamprovides a means to place the laser away from thepart and still minimize the hazard to people. Alsofiber optic light transmission allows for ultrasonicevaluations inside components with access providedby a small hole.

FY 98 5-5

Figure 4. Comparison of signals. The upper waveform wascaptured using the Michelson interferometer with no fiber,while the lower was captured using fiber in the probe arm.Only a slight decrease in signal-to-noise ratio is seen. Theabove are through-transmission on epicenter.

0 100 200 300 400 500 600 700 800 9001000-0.07

-0.06

-0.05

-0.04

-0.03

-0.02

-0.01

0

0.01

0.02

Am

plit

ude

(V)

Laser flash

Longitudinal

Shear

Time (x10 ns)

Am

plit

ude

(V)

Time (x10 ns)0 100 200 300 400 500 600 700 800 9001000

-0.02

-0.015

-0.01

-0.005

0

0.005

0.01

0.015

0.02

Longitudinal

Laser flash

Shear

505 Thomas_qk 7/27/99 8:52 AM Page 5-5


Applications

The laser-based ultrasound laboratory has thecapability to evaluate a wide variety of materials,and has looked at materials including aluminum,stainless steel, uranium and paper. Thus we havedemonstrated the applicability of laser-based ultra-sound on a variety of important, program-criticalmaterials. This capability will allow us to supportprograms at LLNL such as those for weapons,lasers, nonproliferation, and energy. The followingsection briefly describes programmatic applicationsthat resulted from this research.

Process Control. As a sensor for processcontrol, ultrasonics is extremely valuable. The speedat which ultrasonic testing can acquire measure-ments facilitates feedback control for machiningoperations. We selected a process control problemas a vehicle to direct our research. The process wasa cutting application with a state-of-the-art laserthat produces very short pulses. The pulses from thecutting laser generate acoustic signals that containinformation about the cutting process.

Our first challenge was to understand thephenomenon and the types of ultrasonic waves thatare generated. We modeled the system with acomputer code that calculates the expected ultra-sonic signal, based on laser input parameters andthe material properties and geometry. Initial testswere run on thin stainless steel plates.7 The cuttinglaser beam was directed at the plates, and theMichelson interferometer was used to detect theultrasound generated by the cutting beam.

The Michelson interferometer was able to detectboth bulk and surface waves, depending on the testset-up. In one set-up, detection was performed onthe side of the plate opposite that on which thecutting beam was incident, for the capture of bulkwaves. In the other set-up, the cutting beam and theinterferometer beam were both incident on the sameside, for the capture of the surface waves. This config-uration represented the process monitoring set-up.

Next, the Michelson interferometer was used todetect the ultrasound generated by the cutting laserin a real part.5 At the start of the cut, a clear signalwas seen, and this signal disappeared by the timecut-through occurred. This loss of acoustic signal atcut-through is the parameter selected for sensingwhen the laser had cut through the part. Aftersuccessfully detecting the signals generated by thecutting laser with the Michelson interferometer, weswitched to a Fabry-Perot interferometer for ultra-sound detection. Whereas a strip of reflective tape

was required at the detection site when using theMichelson interferometer, no tape is needed for theFabry-Perot, and thus detection on non-specularlyreflecting surfaces is possible.

The case for the process monitoring applicationinvolves non-specularly reflecting components. TheFabry-Perot interferometer is designed to operatewith minimal operator intervention. Light for theinterferometer is transmitted to and from thecutting chamber by optical fibers 20 m in length. Totest the sensitivity, a laser-based ultrasonicthrough-transmission test on an aluminum platewas run using a Nd:YAG laser to generate ultra-sound in the thermoelastic regime and the Fabry-Perot interferometer for detection. The waveformfor this case is shown in Fig. 5.

The Fabry-Perot interferometer was also used todetect laser-generated ultrasound on a stainlesssteel surrogate part, to more nearly mimic the test-ing on actual parts. Figure 6 shows a waveformsignal-averaged from 10 laser pulses. The stainlesssteel part was a thin shell, and the waveform showsa dispersive plate wave. Over the course of the lasercutting project, various parameters of the cuttingpulses such as energy and repetition rate werechanged. A piezoelectric transducer was used toverify that the pulses were still generating ultra-sound once the Fabry-Perot interferometer wasobtained. Once the feasibility was demonstrated, theprogram funded this feedback control sensor as astandby option that can be implemented if needed.

Laser-based Ultrasound Sensor for PaperManufacturing. The DOE has an initiative to reduceenergy consumption in the largest industries in the


0 200 400 600 800 1000-8

-6

-4

-2

0

2

4

6x10-3

Longitudinal

Time (x10 ns)

Am

plit

ude

(V)

Shear

Figure 5. Fabry-Perot-interferometer-detected signal of laser-generated ultrasound. This is a through-transmission onepicenter case in aluminum for a thermoelastic source.

505 Thomas_qk 7/27/99 8:52 AM Page 5-6


United States. Paper manufacturing is one suchindustry. We are collaborating with LawrenceBerkeley Laboratory (LBL) on developing a laserultrasonic sensor. Real time process control affordedby this sensor will increase the quality of the prod-uct and reduce the energy consumption.

The researchers from LBL and LLNL are workingto demonstrate ultrasonic characterization of paperas it is being processed. Plate waves are generatedin the paper with a laser, and a Fabry-Perot interfer-ometer detects the ultrasonic signals. Changes inacoustic velocity and attenuation signify variationsin the paper processing. This project is on-going andmay result in a demonstration at a factory.

Work on Aluminum at Elevated Temperatures.One task of a cooperative research and developmentagreement (CRADA) between the DOE and the “BigThree” automobile manufacturers is to explorenondestructive evaluation techniques to improve thecasting of aluminum. Nondestructive evaluationcould play a crucial role in improving cast parts bydetermining the condition of the molten materialsprior to casting. Impurities in the molten metal causedefective castings. These defective castings increasemanufacturing costs as well as energy consumption.

For example, if it can be determined that aparticular melt contains an excessive amount ofoxide, then that melt would not be used to castparts. Laser-based ultrasound offers a means oftesting metallic materials at elevated temperaturessince it is a non-contact technique, whereas contacttechniques, such as piezoelectric transducer ultra-sound, have an upper limit on temperature, abovewhich they will not work.

We are funded to demonstrate laser generationand detection of ultrasound in molten metal. Afurnace that has a port to pass the laser beam hasbeen installed in the laser-based ultrasonic labora-tory. Impurities in the liquid aluminum will reflectthe acoustic energy that was generated by a laser.The acoustic energy reflected by impurities in themolten material is detected by an interferometer.This system will remotely monitor the quality of rawmaterial feeding the casting process and sense prob-lems before defective parts are produced.

Future Work

Although LLNL funding of laser-based ultrasoundhas produced a working facility for nondestructivecharacterization and evaluation of materials, thereis still much work to be done in this area.Refinement of the computer code that calculates theultrasound generated in materials from laser pulseswill allow model-based processing on more complexparts, such as parts containing defects. Continuedadvances in the signal and image processing aspectof the work will lead to increases in the signal-to-noise ratios of experimentally obtained data, andwill thus allow laser-based ultrasonic testing on awider variety of materials and parts.

Recently, a fair amount of research has beenperformed in the area of higher frequency laser-based ultrasound. An interferometric technique fordetection has been reported,8 with a number ofreferences in laser-generated high-frequency ultra-sound. This includes acoustic frequencies >100 MHzand well into the GHz range. These high frequenciesare required when working with materials whosecharacteristic dimension or defect size is <1 µm.This relatively new area of research has the poten-tial for meeting the high spatial resolution demandsemerging in LLNL program applications.

References

1. Scruby, C. B., and L. E. Drain (1990), “LaserUltrasonics, Techniques and Applications,” AdamHilger, New York, New York.

2. Candy, J. V., G. H. Thomas, D. J. Chinn, and J. B.Spicer (1996), “Laser Ultrasonic Signal Processing: AModel-Reference Approach,” J. Acous. Soc. Am., 100(1).

3. Huber, R. D., D. J. Chinn, G. H. Thomas, J. V. Candy,and J. Spicer (1997), “Model-Based SignalProcessing for Laser Ultrasonic SignalEnhancement,” Review of Progress in QuantitativeNondestructive Evaluation, 16, pp. 757–764, PlenumPress, New York, New York.

FY 98 5-7

800 900 1000 1100 1200 1300 1400 1500 1600-2

-1.5

-1

-0.5

0

0.5

1

1.5

2

2.5x10-3

Am

plit

ude

(V)

Time (x10 ns)

Figure 6. Fabry-Perot-interferometer-detected signal of laser-generated ultrasound in a stainless steel shell. The waveformshows the dispersion of the plate waves, that is, the highfrequencies travel faster than the low frequencies.

505 Thomas_qk 7/27/99 8:52 AM Page 5-7


4. Spicer, J. (1991), “Laser Ultrasonics in FiniteStructures: Comprehensive Modeling with SupportingExperiment,” Ph.D. Thesis, Johns Hopkins University,Baltimore, Maryland.

5. Thomas, G. H., D. J. Chinn, R. D. Huber, J. V. Candy,and J. B. Spicer (1998), “Techniques for EnhancingLaser Ultrasonic Nondestructive Evaluation,”Lawrence Livermore National Laboratory, Livermore,California (UCRL-ID-129207).

6. Johnson, D., and D. Dudgeon (1993), “Array SignalProcessing: Concepts and Techniques,” Prentice-Hall,Princeton, New Jersey.

7. Chinn, D. J., R. D. Huber, D. D. Scott, G. H. Thomas,J. V. Candy, and J. B. Spicer (1997), “New Techniquesin Laser Ultrasonic Testing,” Lawrence LivermoreNational Laboratory, Livermore, California (UCRL-ID-125476).

8. Fiedler, C. J. (1996), “The Interferometric Detectionof Ultrafast Pulses of Laser Generated Ultrasound,”Wright Laboratory (WL-TR-96-4057).


505 Thomas_qk 7/27/99 8:52 AM Page 5-8

ondestructive Evaluation of an Aluminum AlloyUsing Hyperspectral Infrared Imaging


Introduction

The HIRIS is a second-generation Fourier trans-form spectrometer developed at LLNL (seeReference 1 for an excellent discussion of HIRIS’spredecessor). While it has been used primarily inremote sensing applications, the focus of thisproject is to assess its usefulness in nondestructiveevaluation applications.2 The HIRIS consists of aMichelson interferometer, a cryogenically cooled

silicon arsenide (Si:AsBIB) focal plane array, andassociated optics.

Figure 1 is a block diagram of a hyperspectralimaging spectrometer.3 Infrared light enters theinterferometer and is split into two paths by thebeam-splitter. Along one path, a fixed mirror reflectsthe light back through the beam-splitter and ontothe focal plane array. Along the second path, amovable mirror reflects the light back through thebeam-splitter which in turn reflects it onto the focal

FY 98 5-9

Fourier transform spectroscopy has found application in many areas, including atmospheric chem-istry and material characterization. This report describes an investigation into the application of theLawrence Livermore National Laboratory (LLNL) Hyperspectral Infrared Imaging Spectrometer(HIRIS) to the nondestructive evaluation of blocks of aluminum alloy. We describe the HIRIS systemand the aluminum alloys used in the investigation, and the technique used to collect the hyperspec-tral imagery. We discuss the processing required to transform the data into usable form and a tech-nique to analyze the data. We also provide some preliminary results.

Randy S. RobertsDefense Sciences Engineering DivisionElectronics Engineering

Fixed mirror

Movable mirrorObject ofinterest

Focal plane array

Beam-splitter

Figure 1. Schematicdiagram of the HIRISimaging system.Light enters the interferometer fromthe left, and is splitinto two paths bythe beam-splitter:towards a fixed-position mirror, andtowards a movable-position mirror. Themirrors reflect thelight back to thebeam-splitter, whichdirects the light ontothe focal plane array.Translation of themovable mirrorresults in a time-varying interferencepattern on the focalplane array.

TA510 Roberts_qk 7/27/99 8:54 AM Page 5-9


plane array. Note that the optical length for the firstpath is fixed, while the length of the second path canbe adjusted by moving the mirror.

If the light impinging on the focal plane array is inphase, constructive interference results. If the lightis out of phase, destructive interference results. Ifthe movable mirror translates with a fixed velocity, atime-varying interference pattern is formed on thefocal plane array. Since HIRIS uses a focal planearray, a time-series of interference data, called aninterferogram, evolves in each pixel. Fourier trans-formation of all interferograms results in the desiredhyperspectral image data.

The objects of our interest are severalaluminum blocks provided to LLNL by GeneralMotors Corporation (GM). The blocks are made ofan aluminum alloy denoted A356.4 This alloyconsists primarily of aluminum (~92%), silicon(~7%) and trace amounts of copper, magnesium,manganese, iron, and zinc. The alloy has excellentcorrosion resistance and casting properties, goodmachinability and weldability, and high strength. Itis most often used in the automotive and aircraftindustries where high strength and low corrosionare required.

The blocks we received from GM were subjectedto compression-tension cycles until they fractured.Typically, A356 reliably fails after approximately500,000 cycles, but these samples failed afterapproximately 5000 cycles. It is conjectured that thepremature failure was due to the addition of salts,such as potassium chloride and sodium chloride,during processing.

The objective of our investigation is to study thefracture sites in the A356 samples using hyperspec-tral infrared imagery, and if possible, determine thevalidity of this conjecture.

Progress

For these experiments, the HIRIS was recon-figured from a remote sensor to a laboratory instrument. Reconfiguration essentially consistedof installing a collimator lens at the input pointof the interferometer. The focal length of thecollimator lens was 100 mm, and the focal lengthof the condenser lens was 226 mm, thereby yield-ing a system magnification of 2.26. The focalplane array in the HIRIS is a 128 × 128-pixelarray with a pixel size of 75 µm/pixel. The result-ing resolution of the system is thus on the orderof 33 µm/pixel.

To collect the data, the aluminum blocks wereheated with a hot air gun until they reached atemperature in excess of 50 ºC. The blocks werethen held at this temperature for the duration of thedata collection. In addition to the A356 blocks, cali-bration data was collected from a blackbody source.The source was set to 55 ºC for a hot reference, and45 ºC for a cold reference.

A scan in the HIRIS consists of translating themovable mirror at a constant velocity through a fixeddistance. As a point of terminology, the resultingsequence of images produced by a scan is referredto as a raw data cube. The scan time and samplingrate of the interferograms determine the spectralresolution of the hyperspectral data. For theseexperiments, the resulting resolution of the systemwas set to 16 cm-1.

To increase the signal-to-noise ratio of featuresin the spectral data, an ensemble of data cubeswere collected from each object of interest. In ourcase, 32 data cubes were collected from eachA356 sample, and 16 data cubes each werecollected from the hot and cold body sources foreach A356 sample.

A portion of one of the interferograms associatedwith pixel (64, 64) of an A356 sample is shown inFig. 2. Note the structure of the interferogrambetween frames 220 and 580. An image associatedwith this part of the data cube, frame 278, reveals aring-like interference pattern, as illustrated inFig. 3.

Interferograms, while illuminating some signalstructure in the data cube, are difficult to interpret.However, the spectral information contained ininterferograms can be quite revealing. To begin theprocessing, the data cubes were first averaged toincrease the signal-to-noise ratio. As a result of theaveraging, three data cubes were produced for eachcollection: an A356 data cube, a hot calibration datacube, and a cold calibration data cube.


Val

ue

1.4x104

0 200 400 600 800Sample

1.6x104

1.8x104

2.0x104

Figure 2. Plot of an exemplar interferogram from an A356sample. This interferogram was formed in pixel (64, 64), and its range has been limited to reveal the most interesting structure.



The spectral content of a data cube is found byFourier transforming each interferogram in the datacube. To reduce spectral sidelobes, each interfero-gram is multiplied by a triangular data-taperingwindow prior to Fourier transformation. To ease thecomputational burden of subsequent analysis, thefringe portions of the spectra are discarded. Finally,the A356 data is calibrated using the hot and coldreferences.1 The calibrated data cube was thenanalyzed as described below.

Spectral data in a data cube is analyzed using atechnique called Principle Component Analysis,5,6

a powerful analytical tool for analyzing complexdata sets. In essence, data is rotated into a coordi-nate system with some highly desirable propertiessuch as orthogonality and minimum basis restric-tion error. The technique begins by estimating thecorrelation matrix of the data. The eigenvaluesand eigenvectors of the correlation matrix are thencomputed. By themselves, the eigenvectors forman orthogonal coordinate system in which torepresent the data. However, the eigenvectorsassociated with the largest eigenvalues corre-spond to the axes that most efficiently representthe data. By representing the data in the new coor-dinate system, that is, a coordinate system usingthe eigenvectors associated with the largest eigen-values, the most important features of the data aremore readily determined.

The distribution of eigenvalues can reveal agreat amount of information about the structureof the data. A plot of the logarithm of the eigen-values of one of the A356 samples is shown inFig. 4. As illustrated in the figure, the magnitudeof the eigenvalues decreases rapidly over thefirst four or five. From approximately the fifth toperhaps the twentieth eigenvalue, the decrease ismore gentle, but the eigenvalues still have alarge magnitude.

The modes associated with these twenty eigen-values comprise the main structure of the data. Ifthe data were reconstructed from these modes, theresidual error would be minimal. The modes associ-ated with the remaining eigenvalues are not important to the analysis. In particular, the modesfrom twenty to approximately eight-five are associ-ated with system noise. The remaining eigenvaluesbelong to degenerate modes. They are artifacts ofthe processing, and are not useful in the analysis.Once the important features of the data set havebeen determined using eigenanalysis, they can beinterpreted for physical meaning. Indeed, this technique has been used to identify constituentcomponents of mineral specimens.1

Summary

The HIRIS developed at LLNL has found manyuses in remote sensing. This project assessed its usein the nondestructive evaluation of several blocks ofaluminum A356. Hyperspectral data was collectedon the blocks after a slight modification to the HIRISinput optics. Several data sets were collected on thespecimens and calibration sources. The data setswere averaged to increase the signal-to-noise ratio,processed, and Fourier transformed to produce data

FY 98 5-11

Figure 3. Frame 278 from the raw data cube of the A356sample used in Figure 2. Note the concentric ring structure ofthe image formed by the interferograms. The black dots resultfrom pixel imperfections in the focal plane array.

Log

(ei

gen

valu

e)

0

-4

-2

2

4

6

0 20 40 60Mode

80 100

8

Figure 4. Plot of the logarithm of the magnitude of the eigenvalues of an A356 imagery. There are three types ofmodes associated with the eigenvalues: modes related to struc-ture of the data (0 to 20); modes associated with system noise(21 to 85); and degenerate modes produced by the processing. (86 to 92).



cubes of spectral data. Preliminary analysis of thespectral data using Principle Component Analysisreveals interesting structure in the data, but additional analysis is required to fully assess thepotential of this technique.

Acknowledgments

The author would like to thank J. Bixler forcollecting the data, W. Aimonetti for help with theanalysis, and C. Bennett for interesting discussionson hyperspectral infrared imagery.

References

1. Bennett, C. L. (1988), “LIFTERS, The LivermoreImaging FTIR Spectrometer,” Fourier TransformSpectroscopy: 11th International Conference, J. A. deHaseth, ed., American Institute of Physics.

2. Roberts, R. S., W. D. Aimonetti, and J. V. Bixler(1998), “Material Characterization Using aHyperspectral Infrared Imaging Spectrometer,”Proceedings of the IEE Thirty-Second Conference onSignals, Systems, and Computers, Pacific Grove,California, November.

3. Beer, R. (1992), Remote Sensing by FourierTransform Spectrometry, John Wiley and Sons, NewYork, New York.

4. American Society for Metals Reference Book, 2nd ed., 1983.

5. Malinowski, E. (1991), Factor Analysis in Chemistry,2nd ed., John Wiley and Sons, New York, New York.

6. Fukunaga, K. (1990), Introduction to StatisticalPattern Recognition, Academic Press, New York, New York.



n-Situ Identification of Anti-Personnel Mines Using Acoustic Resonant Spectroscopy


Introduction

The detection of buried anti-personnel (AP)mines is a problem of tremendous importance inmany parts of the world. The InternationalCommittee of the Red Cross and Red CrescentSocieties estimates that approximately 110 millionAP mines have been placed in 64 countries worldwide.1 The disruption to daily life caused by APmines in these countries is enormous. For example,over the last fifteen years, land mines inAfghanistan have caused 600,000 casualties (oneout of fifty Afghans).

In addition to causing direct injuries, land minesoften contribute to malnutrition and promotedisease in a population. For example, safe drinkingwater can be denied to a village where there is amere suspicion of a land mine field. The villagemight therefore use unsafe supplies, increasing therisk of dysentery. Similarly, land mines can easilyremove farmland from production, increasing therisk of malnutrition.

Modern AP mines are surprisingly simple, effec-tive, and insidious devices. They tend to be cylindricalin shape, with diameters ranging from 6 to 15 cm,and heights from 3 to 6 cm.2 They often havesimple pressure-sensitive triggers, requiring amass of 3 to 25 kg for detonation. The explosivecharge is typically 50 to 200 gm of TNT. Thisamount of explosive can cause serious injury to anadult, such as the destruction of a foot or leg, andit can kill small children.

Modern mines contain very little metal, and thusare very difficult to locate with conventional metaldetection instruments. Operationally, they areburied in soil to a depth of 1 to 4 cm, and laid out ina variety of patterns, depending on the application.They are very robust devices, and can remain opera-tional for decades after planting.

Conventional demining techniques are time-consuming, labor-intensive and dangerous.2 Due toits thoroughness and effectiveness against mini-mum-metal mines, manual probing is the mostwidely used demining technique in the world.

For clarification, we note that we draw a distinc-tion between humanitarian demining and mine fieldclearing. Mine field clearing refers to a military unitbreaching a mine field. Often, this is accomplishedby a simple linear path through the field. Speed is ofthe essence. Safety and thoroughness of mineremoval are secondary. In contrast, humanitariandemining calls for clearing large areas of terrain,and thoroughness of mine removal is paramount (asis the safety of the demining crew). Speed is asecondary concern.

In manual probing, a deminer pushes a probe intothe ground at a shallow angle. The probe is liftedslightly and extracted. A deminer trained in thistechnique can feel a mine above the probe as it islifted and extracted. To ensure that all AP mines inan area are detected, probing is performed over a3-cm-×-3-cm grid. With this type of search pattern,one deminer can clear approximately one squaremeter of land per day.

FY 98 5-13

We have presented a novel technique for discriminating between anti-personnel (AP) mines and otherburied objects. The technique is based on measuring the acoustic resonant spectrum of a buried object,using a probe to provide an excitation signal and a stand-off radar to detect the object’s response.Several experiments were performed on mine-like objects to assess the potential of the identificationtechnique. One set of experiments found spectral features that might be used as the basis for discrim-ination algorithms. The second set of experiments indicate that the spectra of the objects have a toler-able degree of variability. Taken together, we conclude that it is highly likely that AP mines can bedistinguished from other buried objects using acoustic resonant spectra acquired by a stand-off radar.

Randy S. Roberts and Roger L. PerryDefense Sciences Engineering DivisionElectronics Engineering



Emerging mine detection techniques includeinfrared sensing for wide area detection, olfactoryand chemical sensors for explosive detection, andground-penetrating radar for detecting buriedobjects. Of these techniques, ground-penetratingradar has shown the most promise.

A current project at Lawrence LivermoreNational Laboratory (LLNL) seeks to imageburied objects using micropower impulse radar(MIR) as an imaging sensor. The idea is to rapidlysurvey the top layer of soil for objects that havegeometric properties similar to AP mines. Whensuch an object is detected, it is marked for prob-ing. The application of such a ground-imagingsystem would no doubt increase the efficiency of deminers. However, deminers could easily be inundated by the large number of false alarms produced by shrapnel, debris and other objects that produce images similar to AP mines.

In Situ Identification Technique

The technique described here allows a deminer todiscriminate between mines and other buriedobjects in situ. It thus provides a means to distin-guish AP mines from other buried objects detectedby ground-imaging systems. The technique is basedon acoustic resonance spectroscopy (ARS), a tech-nology that has been successfully applied to theidentification of chemical munitions.3–5

The ARS technique is based on the premise thatobjects of interest, such as AP mines, have charac-teristic resonant frequencies that we can excite anddetect and that are nominally reproducible. Sincethe resonant frequencies of an object are functions

of its geometry and construction materials, it followsthat objects of a similar nature will have similarpatterns of resonant frequencies. Thus, thefrequency response pattern of an object can be usedto identify it.

As mentioned, this technique was used with greatsuccess in the ARS Munition Classification System(MCS) developed at Los Alamos NationalLaboratory. An instrument based on this conceptwas developed, tested, and accepted for use as a verification tool for the 1997 Chemical Weapons Convention.

There are several ways to estimate the acousticresonance spectrum of an object. In the case of anAP mine, the objective is to obtain the spectrum inthe most non-invasive manner possible. Figure 1 isa schematic diagram of the technique.

To begin with, an excitation force spanning thefrequency band containing the resonances is appliedto (or near) the object. The magnitude of the excita-tion force does not need to be large to obtain auseful spectrum. In the case of the ARS-MCS, theexcitation force is provided by a piezoelectric trans-ducer. The induced vibrations on the munition haveamplitudes on the order of 10 nm.

For AP mines, a simple approach is to touch theobject with a probe that produces low amplitudevibrations over the frequency band of interest.Although this requires contact with the object, it isperhaps the best way to excite the object’s resonantfrequencies. And as previously noted, touchingmines with a probe is accepted practice with deminers.

With regard to the form of excitation signal, anarrow-band sinusoid stepped over the frequencyrange of interest has a major benefit: the toneprovides a coherent reference for processing thereceived signal. A coherent reference can be used to increase the signal-to-noise ratio of the returned signal, and can also be used to extract phase information.

To collect the frequency response of the object,a sensor is required to detect the object’s vibra-tions. In general, radar is well suited for the non-invasive sensing of vibrations. In particular, theMIR technology invented at LLNL provides a unique means of non-contact, stand-off vibration sensing.6

Even though the objects of interest are buried,the burial depths are shallow enough that a lowpower radar should have little problem sensing thevibrations. By using localization informationprovided by the LLNL MIR ground-imaging radar, therange gate and other parameters on the vibration


Computer

MIR

Computer

MIR

Figure 1. Schematic of the in-situ anti-personnel mine identification technique. A probe excites the buried object witha stepped-frequency sinusoidal signal. The response of theobject is received by an MIR system, analyzed, and classifiedby the computer.



sensing radar can be adjusted to maximize thesignal-to-noise ratio of the returned signal.

Progress

We have conducted several experiments to assessthe feasibility of the identification technique. Theexperiments consisted of measuring the acousticresonant spectrum of several objects over a limitedfrequency range. The measurements were collectedwith unburied objects to maximize the signal-to-noise ratio.

Four objects of similar geometry were used in theexperiments (Fig. 2). The first object was an inertM14 mine; the second a nondescript mine detectiontarget (MDT); and the third object, serving as debris,was a galvanized pipe covered with duct tape. Afourth object, not pictured, which also served asdebris, was a plastic can containing PlayDough.®

The M14 surrogate is made of a hard plastic,roughly cylindrical in shape and approximately4.3 cm in height, and 5.2 cm in diameter. TheMDT is also made of hard plastic, and is approxi-mately 3.5 cm in height, and has a diameter of7.5 cm. The pipe is 8 cm in height and has a diam-eter of 6 cm., and the PlayDough® a cylindricalplastic can approximately 6.1 cm in diameter and8 cm in height.

The objects were secured in a wooden cradle forthe experiments. The excitation signal was providedby an HP35670A signal generator. That signal wasamplified and applied to a Wilcoxan F4 vibrator. Aprobe approximately 1 ft in length was attached tothe vibrator, and touched to the sides of the objects.(See Figs. 3 and 4 for photographs of the experi-mental arrangement.)

Object vibrations were sensed by a low-powerMIR. The radar was placed 60 mm from the surfacesof the objects, in an orientation perpendicular to thesurface. Both the drive signal from the HP generatorand the return signal from the MIR were digitized ata 40-kHz sampling rate.

Spectra from the objects were collected over thefrequency ranges 50 to 100 Hz, 100 to 200 Hz, and200 to 400 Hz. In each band, the frequency of thedrive signal was stepped over approximately thirtytones, each with a duration of 1.5 s. The responsesignals from the radar were re-sampled to 10 kHz,normalized to the maximum amplitude of the excita-tion signal, and Fourier transformed.

Magnitude plots of the frequency responses in the200 to 400 Hz excitation band are shown in Fig. 5.The upper plot is the response of the M14 AP mine,the middle plot the response of the MDT, and the

lower plot the response of the pipe. Notice that thethree objects produce distinguishable spectra overthis frequency range.

The M14 produces a large response in the 325 to400 Hz region, whereas the responses of the other

FY 98 5-15

Figure 2. Objects used in the preliminary experiments(PlayDough® not shown). On the left is an inert M14 mine; inthe center is a nondescript mine detection target; and on theright is a pipe covered in duct tape.

Figure 3. Experimental apparatus used to measure theacoustic resonant spectrum of surrogate mines and debris. Theobject under inspection is a small pipe covered with duct tape.The MIR system (above the pipe) is typically used for speechanalysis.

Figure 4. Another view of the experimental apparatus. Theprobe is attached to a vibrator that provides an excitationsignal to the object.



two objects are much more subdued. In contrast, the275 to 325 Hz band of the M14 is slightly depressedcompared to the same band in the other objects. Theresponses in the 200 to 275 Hz band are somewhatsimilar in that they all tend to increase as thefrequency increases from 200 to 250 Hz. The minesurrogate and the pipe have similar responses in the275 to 400 Hz bands with a noticeable spike in thepipe’s response at 275 Hz, and null at the samefrequency as the mine surrogate.

The other frequency ranges contained similarspectral features. As an aside, spectra from the canof PlayDough® are uninteresting, containing fewfeatures. Lack of features in the PlayDough® spectrais not unusual since PlayDough® and its plasticcontainer are non-rigid. Most of the excitationenergy is absorbed by the PlayDough®, leaving littleresponse signal.

In addition to the experiments described above,several experiments were performed to assess thevariability of the spectral measurements. The experi-ments consisted of multiple measurements of thespectra of the objects over a 200 to 500 Hz band.The multiple measurements were graphed on a

raster plot for comparison. The spectra showedsome degree of variability, but overall the measure-ments appeared repeatable.

Future Work

Although our preliminary experiments indicatedthe feasibility of this approach to AP mine identifica-tion, several critical issues remain. These issuesinclude the structure of the resonant spectra of APmines and similar objects in the frequency rangeabove 1 kHz, the variability of the resonant spectra,and collecting spectra from buried objects.

We estimate that the resonant frequencies of theM14 mine and MDT occur at approximately 1.2, 4.8,10.8, … , kHz.7 (A similar set of resonant frequen-cies occurs for the pipe and PlayDough.®)

The radar that we used in the first set of experi-ments had an upper frequency range of approxi-mately 1 kHz. As a result, we were unable to investi-gate the region above 1 kHz, where potentiallyunique information about AP mines (that is, theirresonant peaks) is located. Concerning spectralvariability, robust discrimination algorithms depend


150 200 250 300 350 400 4500

200

400

600

800

150 200 250 300 350 400 4500

200

400

600

800

150 200 250 300 350 400 4500

200

400

600

800

Hz

(a)

(b)

(c)

Mag

nit

ude

Figure 5. Magnitude vs frequency plots of the three objects in Figure 1: (a) the M14 AP mine, (b) the mine detection target, and (c) the pipe.



on spectral features that can be reliably extractedfrom resonant spectra. The greater the variability ofthe resonant spectra, the less reliable the spectralfeatures, resulting in poor discrimination. Wecollected some data to assess spectral variability,but a large ensemble of spectral data is required fora full assessment.

Finally, although our experiments have laid thegroundwork for this identification technique, themeasurement of spectra from buried objectsremains to be performed. This set of experimentswould collect spectra from objects similar to theones used here, but buried in a variety of soils, suchas sand, rock, and clay, and under a variety of conditions, dry to very wet.

We conclude this section with thoughts on theautomated pattern classification of spectra and thefuture of algorithm development.

Automated pattern classification is necessary toprovide deminers with confidence in the technique.From the deminer’s point of view, the processingshould produce a binary decision (“yes, the object isa mine,” or “no, the object is not a mine”), alongwith a numerical measure of confidence.

The key to providing such a decision to thedeminer resides in the object’s spectral features. Ifinvariant spectral features can be found, clusteringtechniques can be used for classification. Such wasthe case in the ARS-MCS liquid-solid discriminationalgorithm, where features extracted from the spectra of liquid- and solid-filled munitions groupedinto two clusters, one for solids, and one for liquids.

If the spectral features tend to be variable (dueto factors such as soil conditions and the age ofthe mine), a comparative approach can be used.Such was the case in the ARS-MCS munition classi-fication algorithm, where spectral features ofdifferent chemical munitions (in this case, thefrequencies of resonant peaks) tended to vary. Themunition classification algorithm compared spectral features from the unknown munition totemplates of spectral features built from knownmunitions. (The templates quantified the variabilityof the spectral features.)

While the algorithms developed for the ARS-MCSmay not be directly applicable to the proposedproject, they form a strong foundation for futurealgorithm development.

Acknowledgments

The authors would like to thank J. Holzrichterand G. Burnett for the use of their MIR and dataacquisition systems, and T. Woehrle for the use ofequipment in LLNL’s Modal Analysis Laboratory.

References

1. U.S. Army Communications Electronics Command,Night Vision and Electronic Sensors Directorate,Countermine Division, Humanitarian Demining(1998). http://www.demining.brtrc.com. Also seeDirect and Indirect Consequences of Land Mines on PublicHealth, http://www.demining.brtrc.com/contents.htm.

2. Jane’s Mines and Mine Clearing (1997), C. King, ed.,Jane’s Information Group Inc., Alexandria, Virginia.

3. Roberts, R. S., J. T. Chen, O. A. Vela, and P. S. Lewis(1993), “Munition Classification Using an AcousticResonance Spectroscopic Technique,” Proceedings ofthe Twenty-Seventh Annual Asilomar Conference onSignals, Systems, and Computers, Pacific Grove,California, November 1–3.

4. Roberts, R. S., P. S. Lewis, J. T. Chen, and O. A. Vela(1994), “Techniques for Classifying AcousticResonant Spectra,” Proceedings of the Twenty-EighthAnnual Asilomar Conference on Signals, Systems, andComputers, Pacific Grove, California, October30–November 2.

5. Roberts, R. S., P. S. Lewis, and O. A. Vela (1995), “APattern Recognition Algorithm for the BlindDiscrimination of Liquid- and Solid-Filled Munitions,”Proceedings of the Twenty-Ninth Annual AsilomarConference on Signals, Systems, and Computers,Pacific Grove, California, October 29–November 1.

6. Holzrichter, J. F., G. S. Burnett, L. C. Ng, and W. A.Lea (1998), “Speech Articulator Measurements usingLow Power EM-wave Sensors,” J. Acous. Soc. Am.103 (1) p. 622, January.

7. Skudrzyk, E. (1966), Simple and Complex VibratorySystems, Pennsylvania State University Press,University Park, Pennsylvania.

FY 98 5-17


n Acoustic Technique for the Non-Invasive In-Situ Measurement of Crystal Size and Solution Concentration


Introduction

The National Ignition Facility (NIF) at LawrenceLivermore National Laboratory (LLNL) is currentlydeveloping rapid growth techniques for potassiumdihydrogen phosphate (KDP) single crystals. NIFrequires over six hundred KDP optical components,sized about 0.4 m2, for optical switching andfrequency conversion of the beam line. Rapid boulegrowth rates of approximately 10 to 15 mm/day arenecessary to produce the large number of KDP optical components at a reasonable cost. Whileconventional growth methods would take more thanone year to grow a KDP crystal of the size required,NIF scientists are developing a technique to growthe crystal in approximately eight to ten weeks.

Crystals are grown from a point seed in a highlysupersaturated solution of KDP. The supersaturationlevel of the solution is the primary operational para-meter used to determine the growth rate of the crystal.1 Thermal control of KDP solution is used tokeep supersaturation at desired levels. Fast growthrates of such large crystals requires stringentcontrols on the KDP solution concentration.

Several techniques are commonly used to deter-mine solution concentration in crystal growth. Thecurrent method uses visual measurement of the KDPcrystal size. Given the known solid density of theKDP crystal, the mass of salt in the crystal, removedfrom solution is estimated. Subtraction of this

estimated value from the initial starting salt givesthe solution concentration as a function of time.

This method is rather inaccurate because visualcrystal sizing is accurate to only 1 to 2 mm. Othermethods using density (from buoyancy), refractiveindex, and electrical conductivity are also commonlyused commercially by crystal growers to determinesolution concentration. Generally, solution concen-tration is determined from empirical relationships ofthe measured properties. These commercial tech-niques all require access to the solution. Acousticmeasurements offer the possibility of a non-invasivetechnique for monitoring crystal growth. High-frequency acoustic measurements have the advan-tage of non-invasively obtaining information aboutthe solution concentration and crystal size duringthe growth process.

This work assessed acoustic velocity as a diagnostic for KDP crystal growth. Electrical conductivity measurements for solution concentra-tion are used as a comparison. Acoustic velocitymeasurements are also tested as a means of sizingthe crystal during growth.

Progress

We designed a probe to empirically determine therelationship between acoustic velocity and tempera-ture and concentration of KDP solution. Used fordetermining the empirical relationship, this probe is

FY 98 5-19

We demonstrated the use of acoustic measurements for tracking potassium dihydrogen phosphate(KDP) crystal growth. Both KDP solution concentration and KDP crystal size can be found by usinginformation derived from acoustic wave propagation in the solution. Acoustic measurements showgood correlation to conductivity measurements for KDP solution concentration.

Diane J. Chinn and Paul R. SouzaManufacturing and Materials Engineering DivisionMechanical Engineering

Harry F. RobeyLaser Programs

TA 520 Chinn 7/27/99 8:58 AM Page 5-19


designed to operate invasively in the solution. Afterdetermining the empirical relationships, acousticmeasurements can be made non-invasively fromoutside the solution.

Figure 1 shows a sketch and a photograph of theprobe and the measured acoustic waveform. A 15-MHz ultrasonic transducer mounted on top of theprobe sends a wideband pulse (bandwidth = 55 %)propagating into the probe. Fabricated from fusedsilica, the probe has threaded sides to dispersediffracted waves incident on the sides.

Notched into the probe is a 10-mm gage length.The low coefficient of thermal expansion of fusedsilica ensures that the gage length remains constantthrough the temperature range of KDP supersatura-tion. A pulsed acoustic wave travels roundtrip fromthe transducer, into the probe, through the solutionwithin the gage length and back to the transducerthrough the probe. Travel time of the pulse throughthe KDP solution gives the acoustic velocity. Thewideband 15-MHz ultrasonic pulse permits hightemporal resolution of the acoustic velocity in theKDP solution. The probe can discern changes assmall as 0.25% in acoustic velocity.

In fluids, acoustic velocity is directly related todensity. For example, empirical relationships in sea

water show that acoustic velocity increases linearlywith salinity and as a second-order polynomial withtemperature.2 KDP solution should follow acousticalrelationships similar to those found in sea water.

Figure 2 shows acoustic velocity and conduct-ivity data for different concentrations of KDP solu-tion. Curve fitting indicates that acoustic velocity isrelated to temperature through a second-order polynomial. At a given temperature, velocityincreases with solution concentration. The datashow that acoustic velocity peaks between 60 °Cand 65 °C for all concentrations. From linear regres-sion at T = 60 °C,

ν(C0) = 0.6414C0 + 1.534, (1)

where C0 = solution concentration and ν(C0) = acoustic velocity.

For a fixed concentration, conductivity increaseslinearly with temperature, as expected. At T = 65 °C,

c(C0) = 379.7C0 + 82.9, (2)

where c(C0) = conductivity.Empirical equations derived from curve fitting for

normalized velocity and concentration are shown in


Glass rod

Transducer

a

b

c

Heated water

t

a

Measured acoustic signal

Am

plit

ude

b c

KDP solutionGlass Glass

(a)

(c)

(b)

Mixingbaffle

KDPsolution

Figure 1. Probe andwaveform. Attachedto a 15-MHz trans-ducer, the fused silicaprobe with 10-mmgage length is placedin the KDP solutionto measure acousticvelocity: (a) measure-ment configuration;(b) measuredacoustic signal; and(c) photograph of theprobe. Three reflec-tions from the probeare measured in thesignal. Velocity isfound from thetravel time in theKDP solution.

TA 520 Chinn 7/27/99 8:58 AM Page 5-20


Fig. 3. In Fig. 3a, measured velocity, νmeasured, is normalized to ν(C0) by the equation

, (3)

where ν*(T) is normalized velocity. In Fig. 3b,conductivity is normalized to c(C0) by the equation

, (4)

where c*(T) is normalized conductivity. Conductivity shows better correlation to concen-

tration (R = 0.99976) than velocity (R = 0.9106),though for some applications, the non-intrusivenature of the acoustic technique may be of moreimportance than the small loss in accuracy of themeasurement. Also, the conductivity probe is very

c Tc

c Cmeasured* ( ) = ( )0

v Tv

v Cmeasured* ( ) = ( )0

sensitive to the presence of bubbles on its measurement surface, which strongly affect theconductivity reading.

The empirical equations shown in Fig. 3 areuseful for in-situ continuous monitoring of the solu-tion concentration. Given a temperature, theseequations allow us to relate measured values ofvelocity and conductivity to concentration.

Crystal size in solution is currently measuredvisually with a telescope and a scale mounted to theside of the tank. The resolution currently obtainableis approximately 1 to 2 mm. Acoustic measurementsof crystal size were performed with the configura-tion sketched in Fig. 4. In this configuration, transducers are mounted on the outside of the KDPsolution tank. The arrival time of the reflected wave

FY 98 5-21

Temperature (˚C)

Vel

oci

ty (

mm

/ms)

Temperature (˚C)

Co

nd

ucti

vity

(m

S/cm

)

(a)

(b)

30 40 50 60 70 80 901.62

1.64

1.66

1.68

1.70

1.72

1.74

1.76

1.78

30 40 50 60 70 80 90100

150

200

250

300

C0 = 0.3640C0 = 0.3378

C0 = 0.2033C0 = 0.2442C0 = 0.3058

C0 = 0.1833

C0 = 0.3640C0 = 0.3378

C0 = 0.2033C0 = 0.2442C0 = 0.3058

Figure 2. Acoustic velocity (a) and conductivity (b), measuredat different temperatures and concentrations. For a givenconcentration, velocity follows a second-order polynomial fit,while conductivity increases linearly with temperature.

Temperature t (˚C)

Temperature t (˚C)

(a)

(b)

30 40 50 60 70 80 90

40 50 60 70 80 90

C0 = 0.3640C0 = 0.3378

C0 = 0.2033C0 = 0.2442C0 = 0.3058

C0 = 0.1833

C0 = 0.3640C0 = 0.3378

C0 = 0.2033C0 = 0.2442C0 = 0.3058

No

rmal

ized

vel

oci

ty v

*(T)

0.95

1.3

1.2

1.1

1.0

0.9

0.8

0.7

0.96

0.97

0.98

0.99

1.00

1.01

v*(T) = -0.00001 T 2 + 0.0016 T + 0.9531

No

rmal

ized

co

nd

ucti

vity

c*

(T)

Tc*(T) = 0.01219 + 0.21

R = 0.9106

R = 0.99976

C0 = 0.3640C0 = 0.3378

C0 = 0.2033C0 = 0.2442C0 = 0.3058

Figure 3. Normalized velocity (a) and normalized conductivity(b) derived from empirical equations. Velocity and concentra-tion are normalized for concentration according to Eqs. 1 and2. Velocity follows a second-order polynomial fit with tempera-ture with the maximum velocity occurring at 60 °C.Conductivity fits its empirical equation (R = 0.99976) betterthan velocity (R = 0.9106).

TA 520 Chinn 7/27/99 8:58 AM Page 5-21


from the crystal face is calibrated to the arrival timeof the reference wave. This technique is sensitive toapproximately 0.1-mm changes in crystal size, resulting in a factor of 10 improvement.

The configuration in Fig. 4 shows a non-invasivemethod of measuring solution concentration in addi-tion to crystal size. Acoustic velocity is obtained fromthe reference signal. Given the temperature andacoustic velocity, the solution concentration can bedetermined using the empirical relations in Fig. 3.

Summary

Acoustic measurements have proven to be aviable, continuous, in-process diagnostic techniquefor crystal growth. The technique has no adverseeffect on the solution. Acoustic velocity measure-ments for solution concentration correlate well withelectrical conductivity measurements. This workdeveloped a supplemental tool for the continuousmeasurement of both KDP solution concentrationand crystal size.

Future Work

Because of its ultimate simplicity and minimalcost, visual observation is currently the methodbeing used to determine crystal size during growth.Solution concentration, however, is being measuredonly before and after the crystal growth run. It wouldbe very desirable to have continuous measurementcapability to obtain this quantity throughout thegrowth run.

This work demonstrated the feasibility of usingacoustic measurements for determination of thesolution concentration. Additional work is needed,however, to minimize the cost of a fielded acousticmeasurement system and to provide a more conve-nient operator interface for converting the rawacoustic signals to a direct reading of solutionconcentration.

References

1. Cooper, J. F., and M. F. Singleton (1985), “RapidGrowth of Potassium Dihydrogen PhosphateCrystals,” Lawrence Livermore National Laboratory,Livermore, California (UCRL-91795).

2. Birks, A. S., and R. E. Green (1991), “NondestructiveTesting Handbook, Ultrasonic Testing,” Vol. 7.


Reflectedwave

Reference wave

KDPsolution

Heated water

Crystal

Figure 4. Illustration of acoustic techniques as a non-invasivediagnostic for crystal growth. Solution concentration is foundfrom a reference wave propagating through the solution.Crystal size is derived from the travel time of the reflectedwave from the crystal.

TA 520 Chinn 7/27/99 8:58 AM Page 5-22

icro-X-Ray Computed Tomography for PBX Characterization


Introduction

Current efforts to characterize the weaponsstockpile have produced heightened interest in theaging mechanisms of plastic-bonded explosives(PBX). Structural modeling offers one method ofunderstanding the aging process. Accurate modelingof PBX crystal and binder components requires arealistic 3-D characterization of the microstructure.

X-ray computed tomography (CT) is an advancedimaging technique that provides 3-D characteriza-tion of materials for nondestructive evaluation. In-situ characterization of structures and componentsmakes CT an attractive method for analyzing exist-ing parts and components. Recent developments inhardware have permitted CT imaging with micronspatial resolution.

In addition to geometric characterization, CToffers quantitative information on the linear x-rayattenuation of component materials. Related tosource energy, material density, and chemicalcomposition, x-ray attenuation can be used to iden-tify and characterize different phases in materials.

Progress

We investigated the use of CT for microstructuralcharacterization of PBX. Two CT systems, KCAT andXTM, were used to quantify the structure of a 2-mm-diameter sample of mock PBX. Component materialsof mock PBX do not have the same x-ray attenuationas PBX. However, the microstructure of mock-PBXapproximates PBX.

The x-ray tomographic microscope (XTM) atLawrence Livermore National Laboratory (LLNL)uses beam line 10-2 at the Stanford SynchrotronRadiation Laboratory (SSRL). Located at theStanford Linear Accelerator Laboratory (SLAC), theXTM facility contains a 15-keV monochromaticsynchrotron radiation source tomography system.1,2

The XTM system scanned the mock-PBX sample at5 µm full-width half-maximum (FWHM) of the pointspread function spatial resolution.

Recently developed by the NondestructiveEvaluation (NDE) Section at LLNL, the high-resolution KCAT x-ray tomographic system uses apolychromatic source.3 The mock-PBX sample wasscanned at an energy of 60 keV peak at 10 µmFWHM spatial resolution.

FY 98 5-23

We compared two micro x-ray computed tomography systems for use in the characterization ofplastic-bonded explosive microstructure. This comparison will help guide the development of a tech-nique to generate an accurate structural model of plastic-bonded explosive from nondestructive x-raytomographic scans.

Diane J. Chinn, Jerry J. Haskins, Clinton M. Logan, Dave L. Haupt, and Scott E. GrovesManufacturing and Materials Engineering DivisionMechanical Engineering

John KinneyMaterials Science and Technology DivisionChemistry and Materials Science

Amy WatersDepartment of Material ScienceJohns Hopkins UniversityBaltimore, Maryland

TA525 Chinn_pbx_qk 7/27/99 9:01 AM Page 5-23


Figure 1 shows CT slices of the 2-mm-diametersample from each of the systems. KCAT data isshown in Fig. 1a, XTM data is shown in Fig. 1b. Theslices are representative slices taken at differentlocations on the sample. Because of higher spatialresolution and higher contrast sensitivity, theXTMslice in Fig. 1b provides sharper edges andbetter contrast of the component structures than theKCAT slice.

Histograms of linear attenuation values for each ofthe systems are shown in Figs. 1c and 1d.Attenuation values in the XTM data show four distinctphases in the sample. The KCAT data does not distin-guish separate phases in the attenuation histogram.

Mock PBX consists of four phases: talc, Kel-F,cyanuric acid, and voids. Expected attenuationvalues for these component materials at discreteenergies ranging from 10 to 100 keV are calcu-lated and presented in Fig. 2. For the monochro-matic XTM energy of 15 keV, expected attenuationvalues are 14 cm–1 for talc, 12 cm–1 for Kel-F,3.4 cm–1 for cyanuric acid, and 0 cm–1 for voidspace. These values are based on 100% density.It should be noted that in-situ density of thecomponent materials can deviate substantiallyfrom 100% density. Consequently, in-situ attenua-tion values can be different from the expectedattenuation values.


Histogram of attenuation values in KCAT slice

Attenuation (cm-1)

11001000

900800700600500400300200100

0-100

-0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0

Histogram of attenuation values in XTM slice

Attenuation (cm-1)

1800

1600

1400

1200

1000

800

600

400

200

0

-200-0 5 10 15 20 25 30

(c) (d)

(a) (b)

Figure 1. KCAT (a) and XTM (b) representative slices, showing the difference in the systems. Histograms of attenuation values inKCAT slice (c) and XTM slice (d) show the difference in contrast sensitivity from the two systems.



In Fig. 3, the four phases of the sample from theXTM scan are segmented from a region of interest(ROI) according to attenuation values. Segmentationreveals the morphology of each of the components.Table 1 shows the range of attenuation values foreach of the phases in the ROI. The range of themeasured attenuation values in the scan in Table 1fits well within the range of the expected attenuationvalues for each of the component materials andallows the identification of each of the phases.

Although the KCAT slice in Fig. 1b shows somefeatures of the microstructure, distinct phases arenot readily identifiable in the histogram in Fig. 1c.Unlike the monochromatic XTM source, the poly-chromatic KCAT source energy ranges from 0 to60 keV. Figure 2 shows the range of attenuationvalues for each of the component materials.Contrast sensitivity between the phases is dimin-ished by using a polychromatic source. As a result,

FY 98 5-25

100

10

1

0.110 100Energy (keV)

Att

enua

tio

n c

oef

fici

ent

(cm

-1) Kel-F

Cyanuric acidTalc

CT slice region of interest Phase 1

Phase 2 Phase 3 Phase 4

Figure 3. Segmentation of the XTM slice, showing morphological information of the phases. This segmentation is based on onefeature, linear attenuation. The original XTM data and all segmented phases (indicated by white areas) are as labeled.

Figure 2. Calculated linear attenuation coefficient vs energyfor three of the four components of PBX, plotted for compari-son to the experimental results. Attenuation is calculated fromchemical compositions and density, where Kel-F = H2C8F11C13at 2.02, cyanuric acid = H10C3N3O2 at 2.50, and talc =H2O12Si4Mg3 at 2.78.



In addition, image processing techniques can beused to identify phases. Preliminary processing indi-cates that curve-fitting the attenuation histogram,thereby forcing attenuation values into four phases,may help segment different phases of PBX. Edgeenhancement routines can help sharpen the image,improving characterization of the microstructure.

Acknowledgments

Some of this work was performed at SSRL,which is funded in part by the U.S. DOE/Chemical Sciences.

References

1. Kinney, J. H., and M. C. Nichols (1992), “X-raytomographic microscopy using synchrotron radia-tion,” Annual Reviews of Materials Science, 22,pp. 121–152.

2. Kinney, J. H., D. L. Haupt, M. C. Nichols, T. M.Breunig, G. W. Marshall, and S. J. Marshall (1994),“The x-ray tomographic microscope: 3-D perspectivesof evolving microstructure,” Nuclear Instruments andMethods in Physics Research A, 347, pp. 480–486.

3. Martz, H., J. Haskins, C. Logan, D. Perkins, D.Rikard, R. Roberts, D. Schneberk, S. Sengupta, and E.Updike (1998), “CT characterization of CVIT-II/III(U),” Lawrence Livermore National Laboratory,Livermore, California (UCRL-ID-130261), June.


the phases in the sample are difficult to identifybased on their measured attenuation values.

Summary

Micro-CT was assessed for use in the 3-D charac-terization of PBX microstructure. Comparison of twomicro-CT systems indicates that a monochromaticsource with 5-µm FWHM spatial resolution providessufficient spatial resolution and contrast sensitivityto characterize and identify mock-PBX microstruc-ture. Our initial attempt to use a polychromaticsource at 10 µm spatial resolution allows imaging ofthe microstructure but does not easily permit identi-fication of phases in mock-PBX.

Future Work

The XTM micro-CT system is not readily avail-able to LLNL’s NDE Section and as such is notan easy option in PBX characterization. Thisproject sought to determine the viability of theKCAT system, a system readily available, forPBX characterization.

Several modifications and processing techniquescan be used to improve the sensitivity of the KCATsystem to PBX microstructure. The KCAT sourcespectrum can be modified to a spectrum moreappropriate for distinguishing the materials inPBX. This can be accomplished by reducing thepeak energy and filtering the source. Spatial reso-lution can also be improved. A plan for improvingspatial resolution of KCAT has been developed andawaits funding.

Table 1. Attenuation values in the XTM data.

Attenuation Slice area Size Phase (cm–1) (%) (µm)

1. Talc 12.4–22 11.7 7.5–352. Kel-F 7–12.4 15.6 10.5–1283. Cyanuric acid 0.1–7 71.2 —4. Void 0–0.1 0.7 3.5–30


valuation of an Amorphous Selenium Array for Industrial X-Ray Imaging


FY 98 5-27

The large market for digital x-ray imaging in medicine has driven the development of flat panelimaging devices. We evaluated one such device using the higher energy x rays typical of industrialapplications. This imager is a 360-mm-×-430-mm array using amorphous selenium (amSe). Thedefining characteristic of this technology is that electrons produced by radiation interaction in the Seare directly collected and processed with no intermediate scintillation or other conversion process.Comparing performance to existing technology, we found 1) the modulation transfer function (MTF)remains high as spatial frequencies approach the limit imposed by the 139-µm pixel size; 2) increas-ing x-ray energy degrades spatial resolution performance somewhat, but the amSe imager performswell in spectra as energetic as 450-kV applied potential; 3) the MTF of this amSe imager dropsrapidly with frequency between 0.0 and 0.1 mm–1 for the heavily-filtered spectra used in this study.Secondary radiation transport in material behind the Se is a significant contributor to this attribute;4) required exposure for the spectra tested ranges from 10 to 100 mR. This is 15 to 100 times lessthan Lawrence Livermore National Laboratory (LLNL)’s home-built system using scintillating glassand a charge-coupled device (CCD) camera. Cycle time for CT applications will be dominated by readand refresh times of nearly 1 min. This will allow for great flexibility in using microfocus (low power)sources or large source-to-detector distances. Finally, we found 5) except for issues of peripheralelectronics, which we did not address, Monte Carlo simulations suggest that this amSe imager couldoffer attractive performance at x-ray energies of 3 MeV or more.

Clinton M. Logan, Jerry J. Haskins, Kenneth E. Morales, and Earl O. UpdikeManufacturing and Materials Engineering DivisionMechanical Engineering

James M. Fugina and Anthony D. LavietesDefense Sciences Engineering DivisionElectronics Engineering

Daniel J. SchneberkComputer Applications Science and Engineering DivisionComputations

Gregory J. SchmidIsotope Science DivisionChemistry and Materials Science

Keo SpringerApplied Research Engineering DivisionMechanical Engineering

Peter SoltaniLiberty Technologies, Inc.Conshohocken, Pennsylvania

Kenneth SwartzSterling Diagnostic ImagingNewark, Delaware

TA535 Logan_qk 7/27/99 9:07 AM Page 5-27


Introduction

Sterling Diagnostic Imaging (Newark, Delaware)is introducing a new type of x-ray imager for medicalimaging. It is designed for use with medical x-rayspectra of about 80-keV average energy. It has anactive area (360 mm × 430 mm) equal to standardchest film, with superior speed and spatial resolu-tion. It is packaged into a box not much larger thana film cassette, and packs an impressive 7.5 × 106

pixels with 14-bit data depth. This new imager uses manufacturing technology

similar to that used for flat panel displays. There is avery thin coating of electrical conductor on the radi-ation side of the panel, followed by 500 µm of amor-phous selenium (amSe). Beneath each pixel is acapacitor to store charge and a thin-film transistorfor read-out. A voltage of a few kV is applied acrossthe amSe so that when x rays produce ionization inthe amSe, the resulting charge moves directly alongelectric field lines and is collected in the pixel inwhich the radiation interaction occurs. This isshown schematically in Fig. 1.

This direct collection of radiation-induced chargeis the most important attribute of this new imager. Itconverts x rays to electrons, which are thencollected and integrated without intermediateconversion to visible light. This eliminates the blur-ring that inevitably occurs in scintillator-basedimagers. The technology is scalable to still higherspatial resolution.

The workhorse system configuration for digitalx-ray imaging and computed tomography at LLNL isa scintillating glass that is mirror/lens coupled to aCCD camera. Compared to this technology, weexpect amSe to be more portable and to have1) higher x-ray sensitivity, 2) larger detector area,3) more pixels, 4) higher spatial resolution(expressed relative to the field of view), and 5) feweralignment issues.

There are several unknowns with respect to theapplication of amSe to nondestructive evaluation

(NDE) applications. AmSe has only been demon-strated at relatively low x-ray energy used formedical imaging. Most NDE applications use higher-energy x-ray spectra. More energetic x rays willpenetrate deeper into and through the panelsubstrate to the electronic components behindthe substrate with unknown effects. Also, resolu-tion may degrade at higher x-ray energy as aresult of secondary radiation (x rays and Comptonelectrons) transport.

In addition, dark current in the imager may be anissue with the longer integration times required forindustrial imaging. Finally, x rays must be turned offduring read-out. We need to understand the issuesassociated with computer control of on/off withrespect to our x-ray sources.

Progress

We evaluated 1) spatial resolution, 2) darkcurrent, 3) reproducibility, 4) latent image effects,and 5) dynamic range. In addition, we used the MonteCarlo code GEANT to model radiation transport in thedetector up to MeV x-ray energies relevant to imaginghigh opacity objects such as turbine blades, nuclearweapons, and engine components. This work isintended to suggest what the spatial resolutionperformance might be at MeV energies and indicatepossible solutions to shielding the electronics.

We evaluated possible methods for turning x rayson and off for our 450-kV Philips tube and for the9-MV electron linac.

X-ray Spectra

We acquired data using three different x-ray spec-tra (Fig. 2), filtering to narrow the energy range.The softest x-ray spectrum tested used a tube poten-tial of 120 kV and 76 µm of W filtration; the interme-diate a tube potential of 250 kV and 3.2 mm of Cufiltration; the hardest a tube potential of 450 kV and6.5 mm of Cu filtration.


X rays

–

+

X-rayinteraction

Positive electrode

amSe semiconductor

Negative electrode, capacitor and transistor(7.5 million of them)

Figure 1. Schematicrepresentation of thelayer structure anddetection method ofan amSe imager.


Center for Nondestructive Characteriztion

Source Stability

The amSe imager requires that x rays be absentexcept during the manufacturer-set interval duringwhich the image is acquired. The only method avail-able to us for these tests was to use the preset expo-sure-time-mode of the LLNL Philips model 451 x-raysource controller. This method of control causes thehigh voltage to be ramped up and down to turn xrays on and off. To observe drift or change in x-raysensitivity of the amSe imager, it was necessary toestablish that exposure controlled in this way issufficiently reproducible. Details of this evaluationare given in Reference 1.

To evaluate the reproducibility of exposurecontrolled from the tube controller, we ran a seriesof exposures for both the lowest and the highest kVspectra used for the amSe detector tests. The dataindicated sufficient source stability to do the detec-tor characterization without an independentmeasurement of source output for each image.

Monte Carlo Modeling Results

We used the Monte Carlo code GEANT2 to modelradiation transport in the detector. We chose tomodel monoenergetic x rays at energies of 70, 120,170, 450, 3000 and 9000 keV. The first three ener-gies were selected to approximate the effective ener-gies of the polyenergetic spectra that we testedexperimentally; 450 keV is the maximum x-rayenergy produced in the hardest spectrum we tested(Fig. 2); we chose 3000 keV because it is a techno-logically important energy for large dense objects;and 9000 keV is the end-point of the spectrumproduced by our 9-MV linac.

We calculated several configurations to help usunderstand how the imager might be improved formore energetic x-rays. Results from two configura-tions are presented here. The first configuration isthe array as it exists, while the second explores theeffect of adding a Au layer in front of the Se.

The calculated interaction probability for0.5 mm of Se as a function of photon energy isgiven in Fig. 3. By putting material in front of theSe, it is possible to suppress the response to lowenergy x rays while possibly increasing theresponse to high energy x rays. The potentialincrease at high energy comes from transport ofsecondary radiation (primarily electrons) out ofthe Au and into the Se. To have the desired effect,the material should be dense and high-Z. We usedcomputational Au since it is free.

For these calculations, a monoenergetic x-raysource was made incident normal to the detector

surface along a line. The structure of the amSeimager was represented by 12 layers of variousmaterials. The Se is 0.5 mm thick in the SterlingDiagnostic Imaging design. GEANT transportssecondary photons and electrons. We present herethe results of energy deposition in the Se. Theproblem was zoned so that we could tally energydeposition as a function of distance from thesource. We used 10-µm-thick lateral spacing andextended the calculation to a distance of 10 mmfrom the source. To avoid edge effects, we talliedenergy deposition in a central 3-mm band within a1-cm-wide panel. A schematic of the calculationgeometry is shown in Fig. 4.

FY 98 5-29

0.001

0.010

0.100

1.000

10 100 1000 104

Inte

ract

ion

pro

bab

ility

Photon energy (keV)

Figure 3. Calculated probability of x ray interacting in a layerof Se 0.5 mm thick.

0

0.2

0.4

0.6

0.8

1.0

1.2

0 100 200 300 400 500

120-kV spectrum250-kV spectrum450-kV spectrum

X-ray energy (keV)

Pho

ton

s/eV

•s•p

ixel

Figure 2. Calculated x-ray spectra used to evaluate andcompare imaging systems. Spectra are in units of photons/eV·sinto each pixel for the source-to-detector distance and tubecurrent used with the amSe imager. The 120kV, 250kV, and450kV spectra are integrated over 11, 25, and 37 d photons/s.pixel, respectively.



Figure 5 presents the calculated energy depo-sition for the two configurations reported and thesix different x-ray energies. Energy deposition byprimary and secondary radiation produces elec-trons in the Se that are then collected by theapplied voltage. There may be some variation inrecombination with the ionization density of thesecondary radiation, but, to first order, the signalin the detector should be proportional to theenergy deposited by the incident radiation. Thecalculations indicate that in the present configura-tion, the amSe panel should have a decreasingresponse (per incident x ray) over the energyrange that we tested experimentally.

The location of the minimum at 450-keV x-rayenergy is highly approximate since we calculated asparse energy set.

Addition of the Au gives part of the desired effecton energy deposit. It suppresses the amSe imagerresponse at 70 keV by two orders of magnitude. Theresponse at 450 and 3000 keV is nearly unchanged.For imaging high-Z objects with the LLNL 9-MVlinac, this may result in substantially better imagingperformance. Most of the important imagingphotons are between 1 and 4 MeV. Most of the scat-ter is below 450 keV. It is disappointing that theresponse at 3 MeV is not improved, but we tried onlyone material and thickness. The important conclu-sion is that a tool exists to alter the energy responseof this imager.

There is, however, a potential advantage of thealtered configuration for imaging 3-MeV x rays. Forthe amSe imager in the present configuration, it ispossible that the energy deposit comes from a smallnumber of high-energy events. This could make theimage quantum-noise limited. The configuration withthe Au has much higher interaction probability,possibly leading to more interaction events beingrecorded and therefore to lessened quantum noise.Insight into these effects is available in the MonteCarlo calculations already performed. Time andbudget did not permit this evaluation.

The shape of the energy deposition as distancefrom the (line) source increases is given in Fig. 6 forthe lower energies computed. The zone width in theMonte Carlo calculation was 0.01 mm.

Figure 7 shows the shapes of the energy deposi-tion binned into pixels the size of the amSe imager.The data have been normalized so that each energyhas the same integral. These data suggest that thereshould be an experimentally observable effect on theline-spread function from secondary radiation trans-port as x-ray energy is increased up to a tube volt-age of 450 kV. They also show that the line-spreadfunction will be further broadened by secondary


0.01

0.1

1

10

Ener

gy

dep

osi

ted

(ke

V/i

nci

den

t p

ho

ton

) 100

0 100 1000 104

Energy (keV)

Present commercial configuration

1-mm Au added in front of Se

Figure 5. Calculated energy deposition in Se layer of amSeimager as a function of incident x-ray energy.

10 mm

X-rays

10 mm20 mm

3 mm

Talley zone

Figure 4. Schematic of configuration used for Monte Carlocalculations. Tally zones are within a 3-mm band centered ina 10-mm band.

0

0.2

0.4

0.6

0.8

0 0.05 0.1 0.15 0.2Distance (mm)

Rel

ativ

e en

erg

y d

epo

sit

170 keV120 keV

70 keV

Figure 6. Calculated energy deposition in Se layer of amSeimager as a function of distance from line source. Zones in thecalculation are 0.01 mm wide.



radiation transport at energies produced by LLNL’s9-MV linac.

Figures 6 and 7 do not reveal one very importantaspect of the computational results. A significantportion of the total energy deposited is containedwithin the long tails of the line-spread functions.

Figure 8 presents the computational results forthe configuration of the commercial amSe imageras running integrals. The ordinate is the fractionof the total deposited energy that is containedwithin a distance, x, of the line source. We havenormalized these curves to a value of 1.0 at10 mm from the source.

In the type of display given in the figure, thedesired response is an immediate rise to 1.0 at avery small distance, and then no additional energydeposited at larger distances from the incident radi-ation. This spread represents only the effects ofsecondary radiation transport in the amSe imager. Itdoes not include object scatter, air scatter, elec-tronic crosstalk, source unsharpness, digital pixelsampling and many other things.

We see from Fig. 8 that significant energy isdeposited many pixels from the location of the inci-dent radiation. For example, 10% to 30% of theenergy is deposited more than 1 mm from the inci-dent radiation, even at medically relevant energies.The importance of this can most easily be appreci-ated if one imagines imaging a small, high-opacity,object within an otherwise empty field of view.

Even if no source photons reach the amSeimager in the shadow of the object, the aggregateeffect of inscatter from the other 7.5 million pixelswill cause significant detector response in theshadow. This can severely limit the useful dynamic

range and contrast sensitivity in NDE applications.This same characteristic exists to some extent inevery technology available for quantitative digitalx-ray imaging.

Results for the configuration with 1 mm of Auinserted before the Se are shown in Fig. 9. Thecurves for lower energies are degraded from thecommercial amSe configuration. This is of onlyintellectual interest since the response is sostrongly suppressed. The two most technologicallyimportant cases are 450 keV and 3 MeV. These arecompared in Fig. 10. Clearly the presence of a Auconverter layer in front of the Se will have negligi-ble effect on spatial resolution at 450-keV and 3-MeV x-ray energy.

These Monte Carlo calculations show thatsecondary radiation transport in the imager itselfcreates long tails on the line-spread function of theamSe imager in its commercial configuration for allenergies simulated, 70 keV to 9 MeV.

For imaging with the LLNL 9-MV linac, spatialresolution will be degraded compared to lowerenergies, but potentially better than any otheravailable technology for area imaging at theseenergies. Adding a high-Z converter in front of theSe is a potentially viable method of adjusting theimager x-ray energy response with negligibleimpact on spatial performance.

amSe Image Acquisition Procedure

Initially we had intended to collimate the x-raybeam to approximately one quadrant of the amSeimager. In the event that our tests did irreparabledamage, this strategy would have retained some

FY 98 5-31

0

0.02

0.01

0.04

0.05

0.03

0.06

0.07

0 0.1 0.2 0.3 0.4 0.5Distance (mm)

Rel

ativ

e en

erg

y d

epo

sit

70 keV

120 keV

170 keV

450 keV

9 MeV3 MeV

Figure 7. Calculated energy deposition in Se layer of amSeimager as a function of distance from line source. Data presentedhere are binned to the pixel pitch of the amSe imager.

0

0.2

0.6

0.8

0.4

1.0

1.2

0.01 0.1 1 10Distance from incident photon (mm)

Frac

tio

n o

f en

erg

y d

epo

site

d in

sid

e d

ista

nce

= x

70 keV

120 keV

3 MeV

450 keV

170 keV

9 MeV

Figure 8. Calculated shapes of the integrated energy deposi-tion as a function of distance from the line source for the amSeimager in the commercial configuration.



functionality. However, the imager has an internalcalibration procedure that must be run for eachx-ray spectrum. To complete this calibration it isnecessary to fully illuminate the entire panel. Weremoved all collimation from the source andallowed the x rays to f i l l the entire panel and surroundings.

To provide shielding for peripheral electronicswithin the imager, we applied a 0.25-in. Pb shieldingin front of the wide “borders” of the imager. We haveinsufficient knowledge of internal construction toknow whether this shielding protected the vulnera-ble components. This Pb was installed after the data

set at 120 kV had been acquired, but this is of noconsequence, since 120 kV is within the normaloperating envelop of the amSe imager.

Spectral and exposure data is presented inTable 1. Prior to acquiring the set of images foreach spectrum, we performed the “calibration”procedure as outlined by the manufacturer.

We see from Table 1 that exposure times arequite short. If we were to cut the distance to1.5 m and run at maximum current, exposuretimes would be well under 1.0 s. If we operatedwithout source filtration, it would approach 0.1 s.To make full use of this amSe imager in the


Table 1. Spectral and exposure data for amSe imager.

Tube Filter material Tube Exposure Source-to-voltage and thickness current time detector distance Exposure

(kV) (mm) (mA) (s) (m) (mR)

120 W: 0.076 1.75 10.0 3.00 10.2250 Cu: 3.2 3.00 10.0 3.00 40.8450 Cu: 6.5 2.00 12.0 3.00 102

0

0.2

0.6

0.8

0.4

1.0

1.2

0.01 0.1 1 10

Frac

tio

n o

f en

erg

y d

epo

site

d in

sid

e d

ista

nce

= x

3 MeV450 keV

Denotes Au added cases

Distance from incident photon (mm)

Figure 10. Effect on energy deposit shape from adding 1 mmof Au in front of the Se.

Table 2. Spectral and exposure data for PCAT system.

Tube Filter material Tube Exposure Source-to-voltage and thickness current time detector distance Exposure

(kV) (mm) (mA) (s) (m) (mR)

120 W: 0.076 7.50 110 1.52 1873250 Cu: 3.2 3.50 100 1.52 1854450 Cu: 6.5 2.00 85 1.52 2814

0

0.2

0.6

0.8

0.4

1.0

1.2

0.01 0.1 1 10Distance from incident photon (mm)

Frac

tio

n o

f en

erg

y d

epo

site

d in

sid

e d

ista

nce

= x

70 keV

120 keV

170 keV

3 MeV450 keV

9 MeV

Figure 9. Calculated shapes of the integrated energy deposi-tion as a function of distance from the line source for the amSeimager with 1 mm of Au inserted in front of the Se layer.



future, it is imperative to have some type of shut-ter control over exposure. These exposure timesare less than the time required for the tube toreach operating voltage if we use the tubecontroller as the on/off switch.

Glass CCD Test Procedure

For the comparative images acquired with thePCAT system, we set up PCAT so that the pixel sizewas 103 µm, slightly smaller than the pixels in theamSe imager. This resulted in a 105-mm field ofview. We used the fiber optic scintillator that is140-mm square and 12-mm thick. Parameters aregiven in Table 2.

The objects we imaged are 1) a line-pair gage;2) a Ta cylinder; 3) a Ta edge; and 4) a 3/8-in. sockethead screw with nut (amSe imager only).

The overall configuration of the set-up for dataacquisition is shown in Fig. 11. The amSe imagerwith objects in place is shown in Fig. 12.

With PCAT we omitted the bolt. We imaged a Pbruler to establish the pixel size and field of view. TheTa edge and the cylinder were imaged together. Theline-pair gage was imaged separately.

FY 98 5-33

Figure 11. The overall arrangement used for data acquisitionwith the amSe imager. The Philips tube head with high-voltage cables is on the left. Three meters away on the table is the imager.

Figure 12. Close up of the amSe imager with objects in place.The white foam was used to support some objects. The whitemarks on the imager face denote the active imaging area ofthe device.

Table 3. Exposure advantage, corrected for pixel area.

Tube Filter material AmSe voltage and thickness exposure advantage

(kV) (mm) (X)

120 W: 0.076 100250 Cu: 3.2 25450 Cu: 6.5 15

Experimental Results

Exposure Results. We see from the exposureconditions in Tables 1 and 2 that the amSe imagerreaches the desired digital signal level with consid-erably less exposure than the scintillating glass/CCDsystem of PCAT. We can correct the PCAT exposurevalues to account for the difference in pixel sizebetween the two systems. For optically coupledsystems such as PCAT, the exposure required isinversely proportional to the pixel area (constantnumber of x rays per pixel). The exposure advantagederived from Tables 1 and 2, corrected for pixelarea, is given in Table 3.

Dark Image Results. To evaluate the magnitudeand stability of the dark image from the amSe imager,we manually selected a region 300 × 600 pixels thatcontained few bad pixels. These images were acquiredwith the same integration time used for x-ray images.The mean and standard deviation for the selectedregion are given in Table 4. The last two digits of theimage indicate the minute of image acquisition.

Two of the dark images, 1607 and 1609, havesignificant counts, about 1% and 2% of the usabledynamic range of the imager. We don’t understand



this behavior. We believe that the images written todisk have been dark-corrected by the Sterling soft-ware, so we expect an average of zero, with a vari-ance caused by noise in the dark images.

The images appear to have been clipped sothat no negative values appear. We also cannotrule out the remote possibility that images 1607and 1609 may have been affected by externalinterference. Four of these images are effectivelydead zero with no significant standard deviation.This needs more work.

Temporal/Radiation Stability. One of the prob-lems in our scintillator-based systems is complextemporal behavior. The particular glass formulationthat we use has both a long-lived “afterglow” andlight output that increases with recent radiationexposure. We checked the amSe imager for a resid-ual image within a dark image after a series of five

object images. Some residual image is observable.This is shown in Fig. 13.

The object image at the top left is the fifth in aseries of five acquired using the 120-kV spectrum. Aportion of the line-pair gage is shown along with aline-out in units of digital levels. The lower image isa dark image acquired as soon as our manual proce-dure allowed (~2 min) after the fifth in the object-image sequence. The two images are shown withdifferent gray scales to reveal the residual image.The magnitude of the residual image is quite small,within the variation in overall dark image variationin Table 4. This effect disappears entirely in asecond dark image.

Another form of temporal bad behavior is forthe imager signal to exhibit dependence on priorexposure. To evaluate this, we acquired five blankimages for each x-ray spectrum tested. These wereacquired as rapidly as our clunky manual methodand the amSe imager control software wouldallow, typically one exposure every 2 min. Weexamined a 300-×-600-pixel patch from eachimage and determined mean, standard deviation,median and number of “bad” pixels, where “bad” isdefined as having a digital level >15,700 or <2000.

Mean digital values are plotted in Fig. 14 forthree sequences of f ive blank images, onesequence for each x-ray spectrum. These varia-tions are very large, greater than 10% for the


Table 4. Dark image results.

Image Mean Standard deviation

1605 0.00374 1.381607 70.5 11.61609 171 17.41612 0 01614 0.0001 .0421615 0.0003 0.011

Figure 13. Upper left image is the fifth in a series of five images acquired at 120 kV. The object is a 0.1-mm-thick Pb line-pair gage.Just after acquiring image 5, the dark image, lower left, shows a slight residual image. Line-outs are shown on the right for bothimages. Line outs were taken at exactly the same pixels in the two images.



120-kV spectrum. Nothing about the time orsequence behavior makes sense to the authors. Thesequence at 120 kV starts high and drops through-out the sequence. The others drop, then climb. Thenumber of bad pixels is very small, 0.03% in theworst case. No trends with continued exposure areevident. The source output is reproducible to about1%, so the variation of Fig. 14 must arise in theamSe imager. If this is a change in sensitivity that isspatially invariant, then correction is a simplematter. LLNL software routinely renormalizes CTdata. Further work is needed to assess whether thisvariation has any spatial dependence, which wouldbe a serious problem for CT.

Modulation Transfer Function. The MTF is afrequency-domain description of the spatial resolu-tion of an imaging system or component.3

One derives MTF from the image of an edge.First, a line-out is taken. The derivative of the line-out is the line-spread function. The Fourier trans-form of the line-spread function is the MTF. Oneconsequence of the digital spatial sampling in a digi-tal system is that aliasing can occur if the imagecontains frequencies that exceed one-half thesampling frequency. For the amSe imager with139 µm pixels, this limiting frequency is 3.6 mm–1,called the Nyquist limit.

Experimentally determined MTFs for the amSeimager are given in Fig. 15 for all three spectra.They drop very rapidly at low spatial frequency, thenhold up well as the Nyquist limit is approached. TheMTFs decrease with increasing x-ray spectral energy.

Our Monte Carlo modeling simulatessecondary radiation transport, one of the compo-nents of the experimental MTF of Fig. 15. Recall

that we modeled monoenergetic x rays of 70, 120and 170 keV as representative of the three poly-chromatic spectra. On the basis of relativeenergy deposition at large distances from theincident x ray, the Monte Carlo results indicatethat MTF for the 120-kV spectrum will drop lessrapidly at low spatial frequency than will theMTFs for the two harder spectra. We see thisbehavior in Fig. 15.

Experimental MTFs for the glass/CCD imagerat the three x-ray spectra are given in Fig. 16.The effect of harder x-ray spectra is in the samedirection as for the amSe imager, though ofsmaller magnitude. Comparing Figs. 15 and 16,we see that amSe is inferior at low spatialfrequencies and superior at high spatial frequen-

FY 98 5-35

5000

5500

6000

6500

7000

7500

1 2 3 4 5

Mea

n d

igit

al le

vel

Run #

120-kV mean250-kV mean450-kV mean

Figure 14. Mean digital value for a series of five blank imagestaken for each of the three x-ray spectra used in these experi-ments. Note the plot has non-zero origin.

0

0.2

0.6

0.8

0.4

1.0

0 1.510.5 2 2.5 3 3.5Frequency (mm–1)

Mo

dul

atio

n t

ran

sfer 250-kV spectra

120-kV spectra

450-kV spectra

Figure 15. MTFs for three x-ray spectra for the amSe imager. These MTFs were derived from the images of a Ta edge.

0

0.2

0.6

0.8

0.4

0 1.510.5 2 2.5 3 3.5Frequency (mm–1)

Mo

dul

atio

n t

ran

sfer

250-kV spectra

120-kV spectra

450-kV spectra

1.0

Figure 16. MTFs for three x-ray spectra for the glass/CCD imager.These MTFs were derived from the images of a Ta edge.


MTF at Low Spatial Frequency. The MTF ofthe amSe imager at low spatial frequency generally decreases contrast and messes up quan-titative measurement. This is made clear by ourmeasurements of signal level in the shadow of anopaque cylinder. The signal at the center of thecylinder arises when the MTF of the imager at lowspatial frequencies is not quite unity. This

0

0.2

0.6

0.8

0.4

1.0

0 1.510.5 2 2.5 3 3.5Frequency (mm–1)

Mo

dul

atio

n t

ran

sfer

Measured for amSe image

Theory for 139-µm pixel


cies. This comparison is shown in Fig. 17 for the120-kV spectrum.

The MTF from digital spatial sampling is given bythe expression:4 MTFx = sin x/x. In this expression x = spatial frequency (mm–1) · pixel size (mm) · π.We calculate this quantity for 139-µm pixel size andpresent it along with the measured amSe MTF inFig. 18.


0

0.2

0.6

0.8

0.4

1.0

0 1.510.5 2 2.5 3 3.5Frequency (mm–1)

Mo

dul

atio

n t

ran

sfer

Glass/CCD

amSe imager

Figure 17. Comparison of MTFs for the two imaging systemsusing the 120-kV spectra.

100

1000

10000

0 105 15 20 25 30Distance (mm)

No

rmal

ized

dig

ital

leve

ls

amSe imager Glass/CCD imager

450-kV spectrum

250-kV spectrum

120-kV spectrum

Figure 19.Normalized line-outsof the image of thethick Ta cylinder,with both amSeimager and the LLNLPCAT system using afiber optic scintillat-ing glass lens-coupled to a CCD.The data is split at15 mm with theamSe imager on theleft and theglass/CCD on theright. The digitalcounts have beennormalized so thatthe maximum levelis the same in eachsignal. Counts forglass/CCD systemhave had the darkimage subtracted.Counts for the amSeimager have internaldark image subtrac-tion.

Figure 18. Comparison of measured amSe imager MTF at 120-kV spectrum and theory, for a digital sample size of 139 µm.



attribute is important to NDE applications, especially to CT.

The results of these images for the amSeimager are compared to the LLNL glass/CCDsystem in Fig. 19 containing portions of line outsfrom six images (two imagers and three spectra)of the thick cylinder. Each signal has beennormalized so that they all have the same maxi-mum value. The trend with increasing spectralenergy for the amSe imager is consistent with theMonte Carlo results presented earlier.

Example Images. Images of the Pb line-pair gagetaken with the two imaging systems are presented inFig. 20. These images were acquired at 450-kVapplied potential. Since the Pb thickness in the gageis only 0.1 mm, this is a modest-contrast object.

Source Control

The amSe imager is continuously active andtherefore will not operate as a snapshot cameradevice in a constant x-ray field. X-Ray exposuretime must be controlled with the source or ashutter, and x rays must remain off during read

out, image transfer, and preparation of theimager for the next exposure. To imagine futureuse of this imager for CT, this whole processmust be controlled from the computer used fordata acquisition and motion control.

One component of this project was to understandissues regarding source control so that we could layout a reasonable plan for future use of the amSeimager as the data acquisition component of anNDE CT system.

Though our experiments were not extensive, itappears that the LLNL linacs can be operated in amode amenable to computer control of relativelyshort bursts. The ramp-up and ramp-down timesappear to be immediate, though a closer examina-tion may find subtle differences.

We performed these tests using a pulse burstthat produced about 1 R. At the time these testswere performed, we did not have amSe exposureresults for softer x-ray spectra nor the MonteCarlo calculations. These combined resultssuggest that a 2-s linac burst at 100 Hz may betoo long. More detailed testing will be required topin this down and to evaluate operation at lower

FY 98 5-37

(a)

(b)

Figure 20. Images ofthe Pb line-pair gagein the region of2.0 lp/mm. (a)Acquired with theLLNL PCAT system,using a fiber opticscintillator that ismirror/lens-coupledto a CCD. The onlycorrection applied tothis image is darkimage subtraction.The brightness gradi-ent from left to right,and the pattern noiseare of no conse-quence because theywill disappear oncethe image is ratioedto the blank image.(b) The same regionof the line-pair gage,acquired with theamSe imager. Thetwo images aredifferent size becausethey were acquiredwith different pixelsize and we chose toprint them withconstant pixel dimensions.



frequency and/or shorter pulse burst, but it looks promising.

Conclusion

This amSe imager shows great promise for digitalradiography and CT. We encountered no problems inoperating it in x-ray spectra up to 450-kV appliedvoltage. Table 5 provides a highly simplifiedsummary of performance relative to one LLNL CCD-based configuration at the 450-kV spectrum testedin this work.

Future Work

The next steps toward routine use of the amSeimager for LLNL Programs are:

1. Write LLNL software to control the imager. 2. Understand the dark image anomalies that

we observed. 3. Determine whether the variation we see in

blank images is spatially dependent.4. Test the 450-kV shutter we designed for repro-

ducibility of x-ray illumination.5. Understand the impact of the poor MTF at low

spatial frequency and what mitigationmeasures can be applied.


Table 5. Summary of performance of CCD-based configuration.

Feature amSe CCD Comment

Spatial resolution (MTF at 1.0 mm–1) 0.40 0.15Maximum field of view (cm) 43 30 a1.5-m image cycle time (s) 46 120 b3.0-m image cycle time (s) 64 375 c24.7-mm contrast ratio 27 80 d

a) amSe provides spatial resolution and field of view simultaneously. CCD systems can achieve30 cm field of view, but not simultaneously, with 0.15 MTF at 1 mm–1. LLNL maximum field ofview for CCD system is 15 cm.

b) Sum of exposure and read/refresh time. For Philips 451 source operated at 2.0 mA, 450 kV, with6.5 mm Cu filtration. Source-to-detector distance = 1.5 m.

c) Same as (c), except source-to-detector distance = 3.0 m.d) Contrast ratio = (digital level in open field ÷ digital level at center of 24.7-mm-diameter

opaque disk).

6. Perform the same suite of characterizationmeasurements at the 9-MV linac.

7. Evaluate other LLNL CCD-based configura-tions and commercial detectors based onamorphous Si.

Acknowledgments

J. Mahler (LLNL) made this project possible bycatching a bit of our enthusiasm for this technology.

References

1. Logan, C. M., “Evaluation of an Amorphous SeleniumArray for Industrial X-Ray Imaging,” in press.

2. Brun, R., F. Bruyan, and M. Maire (1987), “GEANT3Users Guide,” DD/EE/84-1, CERN.

3. Bushberg, J. T., J. A. Seibert, E. M. Leidholdt, Jr., andJ. M. Boone (1994), “The Essential Physics ofMedical Imaging,” Williams and Wilkins.

4. Barrett, H. H., and W. Swindell (1981), “RadiologicalImaging,” Academic Press, New York, New York.


ANDMARC Radar Mine Detection


Introduction

Land-mine detection is important for soldiersfacing an unknown threat, as well as for civiliansreturning to their land after a conflict. Besides themilitary threat for our troops overseas, land minesthroughout the world pose an enormous hindranceto economic and environmental stability, and indi-rectly to national security. Policy decisions regardingthe development, stockpiling, use, and clean-up bywarring nations are being carried out through high-level international negotiations, but there are manyproblems that still remain—particularly the millionsof mines left in the ground. As a national laboratorydeveloping science and technologies in the nationalinterest, Lawrence Livermore National Laboratory(LLNL) has undertaken several efforts over theyears to address the land-mine problem. 1-10

The main problem is that roughly one billion minesare still in place, killing large numbers of civilians andchildren daily. They will continue to be a problem untilthey can be easily found and removed. Detection andclearance technologies are under investigation by

several agencies, but none are adequate for allsituations, particularly for humanitarian deminingscenarios. This report describes internal research anddevelopment for a new mine detection strategy thattakes advantage of the Micropower Impulse Radar(MIR) developed at LLNL.

We describe the second year of this project and theplans to continue toward its deployment for usearound the world. We begin with an introduction to themine detection problem and alternate approaches.

Mine Detection Problem

The mine detection problem being addressed inthis project is repeatedly listed “number one” bypeople who clear minefields: finding the small plas-tic mines near the surface to be removed after aregional conflict. Other problems are difficult as well(for example, locating a field containing mines,education and care of locals, clearing vegetation,disposing of the mine once found, and verificationthat a field is cleared), but detecting individualmines is paramount.

FY 98 5-39

This report describes work done in the second year of a three-year project, LANDMARC, for detec-tion of land mines using advanced radar technology. The primary objective of the overall project is tostudy the use of small radar sensors, such as Micropower Impulse Radar (MIR), and complementarysensors, with optimized detection efficiency. Besides detector design and experimental tests, otheraspects of radar mine detection are under study, including electromagnetic modeling, signal process-ing, assisted detection, and parallel processing methodologies. Both military and humanitarian demi-ning goals were set as standards of operation so that, in the end, a feasible mine detection systemcould be designed that would meet the goals and constraints of deminers. A first-prototype systemwas also used to measure detector performance in blind tests. The MIR mine detector has beenshown to effectively detect and locate small plastic anti-personnel land mines (with explosive simu-lant and no detonators) in three different soil types and under adverse conditions. It achieved an 85%overall probability of detection in 18 blind tests, compared with 44% with an Army-standard metaldetector, and roughly half the false alarm rate.

Stephen G. Azevedo, Jeffrey E. Mast, and James M. BraseLaser Engineering DivisionElectronics Engineering

E. Tom RosenburyDefense Sciences Engineering DivisionElectronics Engineering

TA 540 Azevedo 7/27/99 9:10 AM Page 5-39


This problem represents one of the major difficul-ties facing humanitarian demining operations,whose goals are to clear land and save lives. It isalso a problem for U. S. forces conducting peace-keeping operations (PKO) in troubled areas aroundthe world. The current methods to find these minesare unreliable and slow.

Some facts about this problem are given below.1. Anti-personnel (AP) mines are small, have low

metal content (a few grams), and are the mostnumerous. These mines operate indiscrimi-nately, killing or maiming 2000 civilians andchildren per month worldwide—draining theresources not only of their own countries, butalso of relief efforts and eventually the globaleconomy. One of the most important needs of“deminers” (people who remove post-conflictland mines around the world) is for improveddetection of individual minimum-metal APmines. These mines are numerous, small,effective, and nearly invisible to most currenttechnology. Large anti-tank mines aredetectable, and therefore not a problem.

2. Metal detectors and mine probes are used in100% of clearance operations. These tools areprimitive, and cause numerous false detec-tions, thereby slowing the operation. In oneinstance in Cambodia, deminers detected 70mines in one acre of land, but they also found500,000 metal fragments in the same field.This false alarm rate slows operations andcauses fatigue that can lead to accidents.

3. Radar technology has never been used bydeminers. Even though many disjointed effortshave tried to develop radar-based mine detec-tion systems in the last 20 years, none are inuse. This is because radar has been large,heavy, expensive, requires power, and hasbeen susceptible to clutter. The wavelengthsused have been much larger than the APmines, thus mines have been below thediffraction limit.

4. Clutter is the limiting factor in all mine detec-tion technologies. The sources of this clutter aremany – surface effects, mine-like targets, elec-tronic or thermal noise, and natural featuressuch as roots, rocks, and holes. The clear needis for improved methods of reducing the effectsof clutter for the technologies of interest.

Approaches

Many people in the international humanitariandemining field agree that the most expedient way toaddress the mine detection problem is to develop a

hand-held, cheap, rugged, and easy-to-use sub-surface mine detector. To do this involves develop-ment of better detection strategies.

Current technologies are either incomplete anddangerous (for example, metal detectors miss plas-tic mines), indiscriminate and slow (for example,dogs are prone to fatigue), or expensive and hencenot available to those who need them. All sensorslook for some distinguishing feature that differenti-ates the mines from their natural surrounding eitherby emissions, geometrical form, internal voids, size,depth, distance from each other, material, conduc-tivity or thermal signature.

A list of detection technologies was compiled byH. Ehlers and H. G. Kruessen of Stiftung Menschengegen Minen (MgM, The Humanitarian Foundation ofPeople against Land Mines), a Germany-based non-government humanitarian relief organization (webpage: http://www.mgm.com) with help from numer-ous organizations including the U. S. Army. Theydivide the sensor systems into two general types:substance-analyzing sensors and imaging sensors.

The substance-analyzing sensors are as follows:passive/active metal detection, chemical detectors,bio-sensors (animals or animal tissue reacting toexplosive vapors), and nuclear detection (by variousbackscatter principles such as x-ray analysis, ther-mal neutron analysis, fast neutron activation andquadrupole resonance). Some of the most commonones that have been tested are shown in Table 1 inroughly increasing order of cost for the sensor.There are many reference materials for thesesensors and none is a clear choice for all situations;that is, a detector that works well in the desert maynot be good for snowy forested mountains.

Because of their relatively lower cost, only thefirst two sensor systems are in common use. Falsepositives are due to clutter in the environment caus-ing the sensor to alarm, so one might imagine asuite of different types of sensors for each possiblescenario that would allow for cross-checking. Thisapproach clearly drives the cost up for completesystems, making it even less accessible to thepeople who need it. What Table 1 does not show isthe reliability of each system measured in terms ofprobability of detection (PD) and false alarm forevery possible situation.

To assess their performance, receiver operatingcharacteristic (ROC) curves are measured for eachdetector. Schematic plots of ROC curves for AP minesare shown in Fig. 1 for metal detectors and ground-penetrating radars (GPRs). PD is plotted relative tothe number of false alarms per square meter. Theclear goal is to reach nearly 100% detection with nofalse alarms (marked as “Desired” on the plot).




Current systems are nearing this number for anti-tank mines, but published results show that AP minedetection is far from that goal. The U. S. Armycounter-mines and the U. N. humanitarian minedetection goals for PD are noted on the graph (90%and 99.6%, respectively).

The LANDMARC goal, as noted on the graph, is toapproach the Army and U. N. goals in incrementalsteps. The clear objective is to continue to move thecurves farther toward the desired sensor goal.Measures of ROC values for various terrain,

environment, and mine conditions will help to trackprogress and determine the success of an MIR minedetector. Success of this project will be defined ashaving made fair and accurate measures of theROCs for MIR, to determine if radar-based minedetection is feasible for military or humanitarianmine-clearance goals.

Imaging sensors are used with computer equip-ment to enhance the user’s vision to reduce the clut-ter problem. They can be designed for surface mines(passive video, active/passive millimeter-wave

FY 98 5-41

Table 1. Tested mine detection technologies and their capabilities.

Type Detection method Advantages Disadvantages

Probes or sticks Carefully probe into the ground and Inexpensive Very dangerous search for hard materials to remove and slow

Metal detectors Induction coil measures weak fields Sensitive to metal Clutter, cannot caused by metal content mines & firing pin find non-metallic

mines

Dogs Measure trace amounts of HE in air Very sensitive to Subject to HE and trip-wires fatigue and

masking

Ground- Microwave reflectance from dielectric 50-70% PD for Clutter, penetrating radar interfaces non-metallic AP resolution,

expensive

Infrared Finds thermal differences between soil Stand-off to 10 m Contrast and mine (or disturbed soil) difference low,

expensive

Thermal neutron Analyze energy distribution of neutron Very sensitive to Large, expensive,activation backscatter using 10-MeV gammas trace HE slow, hazardous

#FA/m20

00.1

0.2

0.4

0.6

0.8

1.0

0.2 0.3 0.4 0.5

Desired

MIR goal

U.N. Humanitarian (99.6%)U.S. Army (90%)

Current GPRs

Metal detectors

PD

Figure 1. Receiveroperating character-istic curves. Thesecurves are indicativeof trends only andwill differ greatly forvarying conditionssuch as soil type,moisture, mine type,or weather.



radar), to look at temperature differences indisturbed soil (active/passive infrared (IR), airborneLIDAR), and at conductivity differences (GPR). GPRhas been attempted by many groups for over 20years, and MgM identifies three types, as follows:

1. Narrowband GPR (<1 GHz) — suitable fordeep objects (up to 100 m); requires largeexpensive equipment; has depth resolutionworse than 1 m.

2. Wideband GPR (FM-CW up to 3 GHz) — reso-lution up to 10 cm at depths to 2 m (the widefrequency band produces more information);needs excessive computing capacity and smalldistance to ground; expensive and not massproduced.

3. Interferometric GPR — uses synthetic aper-ture radar (SAR) principles with the abovesystems to produce better 3-D resolution;needs expensive and large radar and process-ing hardware.

Very little work has been done in the higherfrequencies (up to 8 or 10 GHz) or with extremelywide bandwidths (many gigahertz). Aside from theexpense and size, GPR has always suffered fromhigh false-alarm rates. So, an integrated radar mine-detection system stands the most chance of wide-spread use if the following problems could be solved:

1. clutter, causing false positives (could tryhigher 3-D resolution, wider bandwidth);

2. size and weight of the system (currently toobig for one person);

3. cost of the sensor; and4. computer and software complexity.The LANDMARC mine detection project is directly

addressing all of these problems with GPR and islaying the groundwork for a detection system thathas the potential to make significant contributionsto humanitarian demining.

MIR Mine Detection System

LLNL has technologies and expertise in electro-magnetic modeling,10,11 radar systems,2,10,12,13 andsignal processing14-18 that can be applied toassess the limits of radar mine detection andimproved techniques for clutter reduction. Thestated goal is for detection of small AP minesnear the surface. LLNL technologies to apply tothis problem are the MIR technologies, electromag-netic modeling and simulation expertise, and imageprocessing methodologies.

MIR has optimal radar properties for AP land-mine detection, but there is still development to bedone to prove its reliability and speed characteris-tics. Since the sensors are orders of magnitude

smaller, lighter, and less expensive than currentGPRs, they can be packed together so that, for thefirst time, radar images of what is below the groundcan be made with a portable device. Making imagesof the routinely sub-surface at optimal radarfrequencies reduces the effect of clutter because theoperator can now discriminate a mine from a rockbased on its shape. The potential for a first-everfully-3-D, real-time imaging system capable ofrevealing imagery of volumes beneath the earth’ssurface would be a dramatic contribution by LLNL.Therefore, the MIR technology can directly improvethe two key performance parameters of mine-clearance efforts: reliability and speed of clearance.

The approach taken in this project has beendirected toward the following five areas:

1. customizing the MIR technology for thelandmine detection problem;

2. developing models of the radar and cluttersignatures;

3. incorporating those models into algorithms foroptimal processing;

4. performing experiments to validate thetheoretical models; and

5. identifying sponsors whose needs are for aradar mine detection system.

Research results are encouraging and have beenattracting the attention of mine clearance experts.The goal—which we are convinced is highly likely tobe achieved—is to transition MIR technology intofielded systems that save human lives and mitigatethe tremendous human suffering (thousands of lostlives and/or limbs) caused by AP land mines.

MIR Technology Overview

MIR technology is an entirely new sensorconcept that is based on high-speed pulsed elec-tronics and was developed in LLNL’s InertialConfinement Fusion program.19-23 The radarexhibits a combination of interesting properties,including wideband operation, very low noise-floor,extremely low power consumption, small size andlow cost, array configurability, and noise-encodedpulse generation. An MIR motion sensor circuit isshown in Fig. 2.

Of the MIR technology family, the swept range-gate rangefinder with transmit/receive cavity-backedmonopole antennas is the sensor used for minedetection. The rangefinder, or variations of it, isused for signal and image processing applicationsbecause the return signal is similar to traditional“A-scan” from other radars. From these data,advanced signal processing and detection methodsextract information about the soil under inspection.




Variations on the basic rangefinder can changefrequency (1 to 4 GHz, nominally), maximum range(0.3 to 10 m possible), and sensitivity. Antennasystems control a number of these parameters andhave standard modular connections on more recent

MIR implementations. Plots of the radar pulse(reflected from a metal plate) and its spectrum areshown in Fig. 3. Notice that the radar response isrelatively flat in the passband of 1 to 4 GHz, andalso that system noise and jitter are low.

Progress

Significant progress has been made in all fiveareas. It has been clear from the beginning that theclutter problem is of major importance to minedetection. That is, traditional GPR signals aresusceptible to false signatures that appear to bemines or, worse, can cause the operator to miss amine. The design of radar systems and processingalgorithms in LANDMARC has been directed towardreducing clutter.

One direction to reduce clutter has been to char-acterize and optimize the radar signatures formines. The current radar specifications are given inTable 2 for a four-element MIR array. Importantgoals are to have very wide band signatures withlow ringing. The MIR signal, reflected from ametal plate, and its spectrum are shown in Fig. 3.From this image, the MIR is shown to be stableand very wideband.

FY 98 5-43

Figure 2. MIR motion sensor—a complete radar system (with-out antennae or 9V-battery).

(b)

(a)

(d)

(c)

10-10 109

109

104

104

Figure 3. MIRrangefinder pulseechoed from a metalplate at 30 cm. (a)100 waveformsstacked in an image;(b) the average of all100 waveforms, (c)and (d) linear andlog spectra of (b),respectively. Theabscissa values are inseconds and Hertz;the ordinate valuesare in arbitrary units.



Besides improving the radar itself, the LAND-MARC project has made an initial design of thesignal processing performed by the system. A criti-cal issue in this design is a clear understanding ofall signal, noise, and clutter sources in the data. Wedefine noise as any contributor to the radarmeasurement that degrades our capability to detectthe targets of interest. With this general definitionthere are many contributors to the system noisemodel, from simple to more complex:

1. thermal receiver noise – generally modeled aswhite, Gaussian additive noise in the radarreceiver.

2. inhomogeneity in the soil – any variations inEM properties of the soil, including densityvariations, moisture changes, buried rocks, ororganic matter, will cause radar reflections.These will be modeled as additive non-Gaussian noise with correlations on alllengths.

3. multiplicative noise – antenna sidelobes andquantization will contribute to noise which issignal-dependent.

4. surface artifacts – because the surface isgenerally the strongest reflector in thereturned radar signal, any imperfections inremoving it can cause artifacts in the residualmeasurements.

5. multiple scattering – in an environment withmultiple strong scatterers such as mines,rocks, or the surface the radar measurementwill contain not only their primary reflectionsbut also multiple reflections caused by themultiple possible paths back to the receiver.These signals will appear as target-likeimages.

We model the noise in the measurement statisti-cally. The noise model can be most generallydescribed by an N-dimensional joint probabilitydensity, where N is the total number of samples inthe measurement. We consider only two aspects ofthis general distribution: the marginal point densityfunction and the correlation of the signal in time(and later in 3-D space).

For the Gaussian case, this provides a completedescription; for the non-Gaussian case, we will allowthe point distribution to vary but still consider onlysecond-order correlations. This limitation may berelaxed later if analysis of test data shows thathigher-order correlations are important.

The development of noise models involves twomain activities: the first is the development of adatabase of measurements from which statisticalmodels can be developed. The noise database mustcover the system and environmental parameterspace described above for signal models. Thesecond activity is the analysis of the noise data toextract models of point statistics and correlations


Table 2. MIR mine detection system specifications.

Parameter Value

Radar Type: Four-element MIR monostatic rangefindersFrequency range 1.3 to 4.2 GHz (3dB bandwidth)Avg. transmitted power 0.1 mWPeak transmitted power 0.2 WPulse repetition frequency 5 MHzPulse rise time 80 ps (10 to 90% of peak)Receiver noise figure 3 to 5 dBPulse width 1 ns FWHMTiming jitter 50 ps p-pInterface Electronics:Timing Rangefinder ramp, controlled by computerRamp frequency 160 Hz (40 Hz for all four)Number of averages 120Number of range bins 256Antenna Type: Wideband ridged hornFrequency range 1 to 18 GHzGain 3 to 10 dBi , highly-dependent on ground coupling geometry (6 nominal)VSWR 1:1.8Computer System: Dual-processor Pentium, 300 MHz



followed by the development of noise simulationsbased on these statistical models.

The signal processing chain, shown in Fig. 4,shows the planned steps for reducing clutter basedon models of the noise components. The generalapproach to detection is based on the developmentof both target and noise models. These models areused to compute a likelihood that the measurementwas drawn either from the signal-present (H1) distri-bution or from the signal-absent (H0) distribution.The models will generally be adapted to themeasured data and generated in real time.

In the 3-D imaging approach to mine detection, asingle-aperture radar can be scanned to form asynthetic 2-D array. An image in x-y-z is formedcomputationally using the real array in the x direc-tion and by a synthetic aperture in the y direction.This 3-D image is then analyzed to detect the pres-ence or absence of mines.

The general form of the detection model for the3-D imaging radar is the same as that of the 1-Dsingle-aperture radar. The main difference is theaddition of the image formation steps, which gainresolution in x-y and also bring in new noisesources. The detection approaches can be expandedto incorporate x-y information into the templatecatalogs to supplement the time-frequency informa-tion in the 1-D approach.

The image formation step is likened to a refocus-ing of the data, and thus demonstrates significantperformance improvements with this additionalinformation. Our principal approach to clutterrejection requires the use of 3-D radar images todistinguish the shapes of mines from naturalsubsurface inhomogeneities.

What we have found is that if the target object(the mine) is smaller than one quarter of the highestwavelength, reconstructive imaging provides littleimprovement in detection. This was verified in fieldtests as well as simulations. The basic signal modeldiscussed above must include the 3-D point-spreadfunction (PSF) of the imaging formation technique.This PSF will limit the resolution of the imagingsystem and can be a very complicated function ofthe radar system, the imaging geometry, and thetarget itself. For our initial analyses, we haveassumed the PSF is position-invariant and consistsof a specific 3-D form convolved with the target andclutter distribution.

More complex effects are being analyzed usingfinite-difference time-domain (FDTD) models andincorporated into the statistical noise models. AnEM model of the MIR signal and its response fromthe ground and various objects (mines) has beendeveloped. The model has been verified by experi-mental means and used to characterize clutter

FY 98 5-45

Noisereduction

filter

Polarimetricmatched

filter

Surfacereflectionremoval

Time-freq.feature

extraction

Imagegeneration

Sequentialdetector

Geometriccorrection

Audiogenerator

Imagegenerator

Energyestimator

Polarimetricresponse

model

Surfacereflection

model

Cluttermodel

Cluttermodel

Mine responsetemplatecatalog

Linearcorrelator

Linearcorrelator

Linearcorrelator

Figure 4. Signal processing block diagram for the LANDMARC mine detection system.



sources in actual data. One of the key results of themodeling effort is that the radar bandwidth for minedetection of near-surface AP mines is 0.5 to10 GHz. This spectrum is higher than thecurrent MIR (1 to 4 GHz).

On-going development efforts are working towardthat end. Further refinements in the radar antennaeand electronics have produced mine detection capa-bilities that are beyond the known state of radartechnology. Algorithm refinements, particularly forremoval of the first-surface clutter source, haveprovided step-by-step improvement in the signal-to-clutter ratio.

Another aspect of 3-D radar imaging is the sensi-tivity of the imaging to motion errors in the radarposition estimates. The analysis of radar imagingparameters showed that precise positioning of radartransceivers is necessary to produce clear imagesthat aid in clutter removal. Our initial design of asystem that meets this requirement, yet is portableenough to use in the field, is a one-meter-squaredarray design shown in Fig. 5. The sensor array headis cantilevered over the area to be scanned and itproduces overhead 3-D images of the sub-surface sothat the operator can “see” not only the location ofthe buried object, but also its depth and shape.After identifying the mines in the image, thesystem positions its sensors over the mines formarking or removal.

Recent experiments with MIR show promise forimproved AP mine detection. The developmentalprototype of the mine detection system, with soft-ware for noise and clutter reduction, was assem-bled. A series of tests were performed at LLNL thatwere intended to challenge the MIR detection capa-bilities and provide realistic statistics for PD andfalse alarms.

The experiments involved unknown locations ofmines (blind tests), using American-made M14mines and intentional clutter in three soil types andin generally wet weather. Combined results of 18independent tests of MIR compared to the SchiebelPSS-12 metal detector (commonly used at this timein most areas of the world) are shown in Table 3.From these results we believe that our approach isjustified and should be continued.

Using laboratory prototypes and software, MIRhas proven itself capable of detecting many plasticAP mine types buried in various soils. From thisbasic capability we plan to develop full mine detec-tion systems that are based on the MIR concept; thatis, radar-based mine detectors that are small, light-weight, low-power (battery-operated), inexpensiveto replicate, and reliable.

This means that the laboratory prototypes wenow have must be packaged, ruggedized, and madeinvariant to temperature or environmental effects.Software for control, processing, and display mustbe optimized and ported to a small (embedded) plat-form. With industry experts, other sensors must beincorporated into the system to further improvedetection probability while minimizing the falsealarm rate.

Extensive field testing must be done for variousmines, soil conditions, and weather. Once the R&Dis completed, we estimate that replication costs ofsuch systems could be below $10,000 when inte-grated by an industry partner.

Summary

The LANDMARC goals for the second year havebeen completed and preparations are underway forcontinuation into FY-99. An array of MIR trans-ceivers has been built, mounted on a portable plat-form, and shown to detect buried plastic AP land


Figure 5. (a) Operational scenario for a proposed MIR minedetection system. (b) Brass-board prototype system for devel-opmental testing.

(a)

(b)



mines using imaging software and off-line process-ing. Results showed this system to out-perform themetal detectors in common use today. The nextsteps are to further refine the system for field oper-ations and to move toward deployment withindustrial participation.

A summary of what MIR currently offers isgiven below.

1. MIR is the smallest, lightest, lowest powerradar system known, and hence cheapest toreplicate by factors of perhaps 100; however,its performance in the field under many condi-tions still needs to be proven.

2. Because it emits a wider frequency band thanprevious GPR systems, the MIR sensor alonewill generally provide greater information.Depth of penetration is enough for most minedetection applications.

3. Unique 3-D imaging software developed forMIR produces underground views not currentlyused in mine detection. It requires accurateposition information and the latest computertechnology; both are achievable and could bedeveloped. The imaging provides either hori-zontal slice planes or a 3-D perspective viewas though peering into the ground. The arrayconcept initiated out in this project couldprovide a rapid acquisition mode for this typeof imaging.

4. The miniature properties of MIR make it cost-effective for arrays. MIR imaging arrays forvehicles, robots, and even hand-held systems,should improve discrimination and speed, butthis needs to be verified and we are only nowstarting full field testing.

These characteristics are very appealing to thoseinterested in mine detection. However, there is stillmuch work to be done. For example, Army person-nel, and others who evaluate mine detection tech-nologies for deployment, require controlledmeasures of PD, false alarm rate, and scan rate intechnology demonstrations. Evaluation of minedetection systems is performed by the Army infields of their own design. These demonstrationsare intended to be part of Phase 3 experiments toevaluate the system. Modeling, radar and software

refinement, and transition to deployment are alsoplanned for the next phase.

Future Work

Recall that the overall objective of the entireproject is to transition the successful MIR technol-ogy into a prototype land-mine detection system thathelps mitigate the tremendous human sufferingcaused by small plastic AP mines. Attainment of thisgoal and wide dissemination of this technology willhelp stabilize local economies and environment, andindirectly enhance global security. Military deminingobjectives will also be addressed by the samesystem. Successful completion of this project willdemonstrate improved mine detection over currenttechnologies, enhance LLNL technical capabilities inradar imaging for other applications, and alsoattract cooperative participation from the U. S. Armyand other sources.

Specific steps to be taken in the next phase areas follows:

1. Incorporate a higher-frequency radar into themine detection prototype. The goal is todevelop a radar with an expected 9-GHz pass-band (at 3 dB). A wider bandwidth will signifi-cantly improve the radar imaging capability.

2. Refine the noise and clutter models to reflectthe wider passband and predict performancein various conditions.

3. Apply the improved models to the algorithmsfor processing, reconstruction, and detection.In particular, use first-surface echo removalto reduce clutter and study super-resolutionmethods through the University of Californiaat Davis.

4. Perform the series of experiments necessaryto validate the above models and measure thedetection performance improvement.

5. Pursue a path toward commercialization.We also plan to demonstrate the system for vari-

ous organizations that have already expressed inter-est in these sensors, including industry, governmentagencies (DOD, State Department, DARPA), foreigngovernments and institutions (Japan, Norway), andhumanitarian relief organizations (World Bank, U. N.).

FY 98 5-47

Table 3. Results of blind experiments of mine detection using a metal detector vs results from MIR.

Average of 18 tests Detection rate False alarms(per detected mine)

AN/PSS-12 (metal detector) 44% 7.95MIR (radar detector) 85% 10.02



The technologies developed for this project willhave utility well beyond the life of this project. LLNLand other federal agencies have a need for radarimaging systems that have similar characteristics.Bridge inspection for the Federal HighwaysAdministration (with whom a 64-element array hasalready been developed) will benefit from the minedetection technologies.

Other imaging for nondestructive evaluation ofconcrete, wood, and other materials is also under-way. The experience and technology gained throughthe MIR mine detection project will significantlyenhance the understanding of radar imaging and willbenefit many new areas.

Acknowledgments

The authors acknowledge the contributions of thescientists, engineers and technicians involved withthe MIR Project. These include J. Brase, M. Bujak,R. Cavitt, G. Dallum, R. Gilliam, G. Governo,H. Jones, M-L. Liu, J. Mast, S. Nelson, T. Rosenbury,R. Stever, M. Vigars, and P. Welsh. Administrativehelp from B. Wooton, M. McInnis, and R. Sachauis appreciated. We also acknowledge the contin-ued support from the Laser Programs Directorate(J. Brase, S. Vaidya, and E. M. Campbell) and theElectronics Engineering Department (S. Dimolitsasand R. Twogood). The AN/PSS-12 metal detector waslent to us by P. Harben of LLNL.

References

1. Anderson, C., W. Aimonetti, M. Barth, M. Buhl, N.Bull, M. Carter, G. Clark, D. Fields, S. Fulkerson, R.Kane, C. Lee, F. Lee, B. McKinley, J. Page, F. Roeske,Jr., T. Rossow, P. Sargis, J. Scarafiotti, P. Schaich, S.Sengupta, and R. Sherwood, (1994), “LLNL Electro-optical Mine Detection Program,” LawrenceLivermore National Laboratory Internal Report,Livermore, California (UCRL-ID-118672).

2. Azevedo, S. G., D. T. Gavel, J. E. Mast, and J. P.Warhus, (1995), “Statement of Capabilities:Micropower Impulse Radar (MIR) Technology Appliedto Mine Detection and Imaging,” Lawrence LivermoreNational Laboratory, Livermore, California (UCRL-ID-5366).

3. Azevedo, S. G., D. T. Gavel, J. E. Mast, E. T.Rosenbury, and J. P. Warhus (1996), “Arrays ofMicropower Impulse Radar (MIR) sensors forSubsurface Detection,” Proceedings of the EURELConference on the Detection of Abandoned LandMines, IEE Conf. Pub. No. 431, Edinburgh,Scotland, UK.

4. Azevedo, S. G., D. T. Gavel, J. E. Mast, and J. P.Warhus (1995), “Land Mine Detection and ImagingUsing Micropower Impulse Radar (MIR),”Proceedings of the Workshop on Anti-personnel MineDetection and Removal, Lausanne, Switzerland,pp. 48–51.

5. Clark, G. A., J. E. Hernandez, S. K. Sengupta, R. J.Sherwood, P. C. Schaich, M. R. Buhl, R. J. Kane, M. J.Barth, and N. K. DelGrande (1992), “Computer Visionand Sensor Fusion for Detecting Buried Objects,”IEEE Proceedings of the 26th Asilomar Conferenceon Signals, Systems and Computers, Pacific Grove,California, p. 466.

6. Clark, G. A., S. K. Sengupta, W. D. Aimonetti, F.Roeske, J. G. Donetti, D. J. Fields, R. J. Sherwood,and P. C. Schaich (1995), “Computer Vision andSensor Fusion for Detecting Buried Objects,”Proceedings of the Symposium of AutonomousSystems in Mine Countermeasures, Monterey,California, April 4-7.

7. Eimerl, D. (1997), “Land-Mine Policy in the Near-Term: A Framework for Technology Analysis andAction,” Center for Global Security Research,Lawrence Livermore National Laboratory, Livermore,California (UCRL-ID-130849, CGSR-98-002), August.

8. Gavel, D. T., J. E. Mast, J. Warhus, and S. G. Azevedo(1995), “An Impulse Radar Array for Detecting LandMines,” Proceedings of the Autonomous Vehicles inMine Countermeasures Symposium, Monterey,California, pp. 6–112.

9. Sargis, P.D., F.D. Lee, E. S. Fulkerson, B.J. McKinley,and W.D. Aimonetti (1994), “Ground-PenetratingRadar for Buried Mine Detection,” SPIE Vol. 2217,Aerial Surveillance Sensing, including Obscured andUnderground Object Detection, April, pp. 4-6.

10. Warrick, A. L., S. G. Azevedo, and J. E. Mast, (1998),“Prediction of Buried Mine-like Target RadarSignatures using Wideband ElectromagneticModeling,” Proceedings of SPIE Conference onDetection and Remediation Technologies for Minesand Mine-Like Targets III, 3392. pp. 776-783.

11. Nelson, S. D., (1994), “Electromagnetic Modeling forTarget-Rich Embedded Environments,” EngineeringResearch, Development, and Technology, LawrenceLivermore National Laboratory, Livermore, California(UCRL-53868-93), March.

12. Azevedo, S. G., J. E. Mast, S. D. Nelson, E. T.Rosenbury, H. E. Jones,T. E. McEwan, D. J.Mullenhoff, R. E. Hugenberger, R. D. Stever, J. P.Warhus, and M. G. Wieting (1996), “HERMES: Ahigh-speed radar imaging system for inspection ofbridge decks,” Nondestructive Evaluation Techniquesfor Aging Infrastructure and Manufacturing, SPIEVol. 2946, pp. 195-204.




13. Warhus, J. P., J. E. Mast, E. M. Johansson, and S. D.Nelson (1994), “Advanced Ground-PenetratingRadar,” Advanced Microwave and Millimeter WaveDetectors, SPIE Vol. 2275, pp. 22-24.

14. Mast, J. E. (1993), Microwave Pulse-Echo RadarImaging for the Nondestructive Evaluation of CivilStructures, Ph.D. thesis, University of Illinois atUrbana-Champaign.

15. Mast, J. (1998), “SIGOP Manual,” LawrenceLivermore National Laboratory Internal Report, (tobe published).

16. Mast, J. (1998), “GPR Manual,” Lawrence LivermoreNational Laboratory Internal Report, (to be published).

17. Mast, J. E., and S. G. Azevedo (1995), “Applicationsof Micropower Impulse Radar to NondestructiveEvaluation,” Engineering Research, Development,and Technology, Lawrence Livermore NationalLaboratory, Livermore, California (UCRL-53868-95), March.

18. Mast, J. E., and E. M. Johansson (1994), “Three-Dimensional Ground-Penetrating Radar ImagingUsing Multi-Frequency Diffraction Tomography,”Advanced Microwave and Millimeter Wave Detectors,SPIE Vol. 2275, pp. 25-26.

19. Azevedo, S. G., T. E. McEwan, and J. P. Warhus(1996), “Microradar Development,” EngineeringResearch, Development and Technology: Thrust AreaReport, Lawrence Livermore National Laboratory,Livermore, California (UCRL-53868-95), pp. 6-17.

20. Azevedo, S. G., and T. E. McEwan (1996), “ModularMIR,” Engineering Research, Development, andTechnology, Lawrence Livermore NationalLaboratory, Livermore, California (UCRL-53868-96).

21. Azevedo, S. G., and T. E. McEwan (1996),“Micropower Impulse Radar,” Science andTechnology Review, Lawrence Livermore NationalLaboratory, Livermore, California (UCRL-52000-96-1/2), pp. 16-29.

22. “The Pocket Radar Revolution,” New Scientist(August 1995).

23. “Radar on a Chip: 101 Uses in your Life,” PopularScience (March 1995).

FY 98 5-49


MAN-3D: A Software Tool-Kit for 3-D Image Analysis


Introduction

Interpretive signal and image analysis plays a keyrole in nondestructive evaluation (NDE). Especiallyimportant are the analyses of 3-D images resultingfrom the use of such diverse probes as x rays, ultra-sound and NMR. Typically, the data consist of a stackof 2-D image slices with each slice representing acertain physical measurement of a horizontal sectionof the 3-D object under investigation. Researchers areinterested in locating, as well as identifying any“anomalies” or “defects” present in the specimen. Weare also interested in quantifying (finding the numberand size, for example) these anomalies or defects.

As a result, we are led to various questions relatedto the 3-D images of the objects under study. Some ofthese questions are negotiable in a straightforwardmanner akin to those involving 2-D imagery. In a 3-Dcontext however, there are questions that are notdirectly answered by the methods used for the 2-Dimages. In some cases, the extensions, when possi-ble, are computationally expensive and must beappropriately modified to be practically applicable.

This is true, for example, for the notion ofconnectedness that plays a significant role in imagesegmentation. In the 2-D case, we have 4-, 6- and 8-connectedness. These notions are well defined.1,2

But how does one define connectedness in a 3-Dimage? If k is the number of voxels adjacent to agiven voxel, a k-connectedness can be defined for kequaling a number ranging from 6 to 26, and subjectperhaps to some 3-D symmetry consideration.Choosing k consistent with a given application isoften a nontrivial task.

For another example, second-order statistics(joint distribution of 2 spatially separated pixels)plays a crucial role in texture discrimination of a 2-Dimage. If we try to extend it in a natural way to the3-D case we are led to second-, or perhaps “third-order statistics” involving, respectively, the jointdistribution of 2 or 3 spatially separated voxels.

Computational complexity, however, precludesthe blind use of third-order statistics even when werestrict our consideration to a spatial separation of1 voxel from one another. Even in the second-ordercase, one has to be selective in limiting the choice ofthe direction and magnitude in the choice of thevoxel separation.

The examples above point to the need for a 3-Ddata analysis paradigm that relies on algorithmsthat are computationally “affordable” and application-specific in scope. The software tool-kit described inthis work is based on such a paradigm. As a result,image-processing algorithms that are “global” havebeen preferred over the “local” ones whenever feasi-ble, often resulting in a faster turn around time.

The software is written in the language IDL3,which is a C-like language providing access tosubroutines that allow many fast and useful image-processing tasks.

In the next section we describe progress in thedevelopment of the tool-kit with reference to thespecific problems in 3-D image analysis that havebeen addressed in this work. We also discuss a fewapplications from projects of recent interest atLawrence Livermore National Laboratory.

FY 98 5-51

In this project a software tool-kit is written to facilitate 3-D image data analysis. It consists ofsubroutines written in the commercially available image processing language IDL. The emphasis is onthree aspects of image analysis: detection, identification and quantification of regions of interest inimage data. All three are important topics in the analysis of 3-D data arising in nondestructive evalu-ation, including data from x-ray computed tomography and from ultrasonic, infrared, radar, NMR andother scans.

Sailes K. SenguptaElectronics Engineering Technologies DivisionElectronics Engineering

TA545 Sengupta_qk 7/27/99 9:15 AM Page 5-51


Progress

A NDE procedure frequently results in 3-D digi-tized images to be analyzed and interpreted. Amongthe basic tasks related to such analyses are detec-tion, identification and quantification of anomaliesor defects, or more generally “regions of interest”(ROI) in the item probed. The detection of a ROItypically involves the detection of a pattern ofpixels/voxels with characteristics that may be eitherknown or unknown.

For example, the detection of regions of a givenshape (rectangles, circles, or spheroids in 3-D), orknown dynamic range of gray levels, or knowntextural characteristics in an image, is a detection ofthe first kind. On the other hand, the detection of allanomalous regions in an image where thepixels/voxels differ significantly from those in thebackground (regional or global) is a detection of thesecond kind.

In the subsection on detection we have concen-trated on detection of unknown or partially knownROIs, relegating the treatment of the first kind to thesubsection on identification. We will see there thatthis task is handled by some traditional methods of

classification/pattern recognition where classifiersare trained to learn the image characteristics of theROIs based on training samples of known cate-gories. Later, an unknown sample is presented tothe trained classifier for classification.

Once the ROIs are detected and identified, weneed to study them quantitatively.

For example, say we have localized and identifiedthe types of grains of certain known textural charac-teristics in an image. The questions include: What isthe size distribution of this particular grain type in theimage? Is this correlated with the size distribution ofanother grain type in the image? What is the extent ofthis correlation, if any? Questions of this type havebeen addressed in the subsection on quantification.

A summary of the main subroutines developed inthis project is given in Table 1.

Detection

As indicated above, detection usually involveslocalization of the ROIs when they exist. There arevarious means of achieving this. We have dealt withsome of these methods, described briefly in thefollowing paragraphs.


Table 1. Partial listing of main subroutines in the tool-kit for 3-D image analysis.

Detection:Thresholding: histo_analyze.pro histo_seg.pro

threshold.pro

Clustering and Neighborhood Statistics: clust_wts.pro* cluster.pro*nabor2.pro nabor2d.pronabor3.pro nabor3d.pronabor3new.pro nabor_mean_range.pronabor_stat.pro

CFAR Algorithm: cfar_1d.pro cfar3d.proSegmentation by Region Merging: reg_merge.pro reg_merge3d.pro Connected Component Analysis: conn_comp.pro conn_comp3d.proSegmentation by Watershed (Meyer): ws_meyer.pro ws_marker.pro

Identification:Neighborhood Statistics: See Detection Shape and Size Features: See GranulometryFeature Subset: See directory “Classify”Classification Routine: See directory “Classify”

Granulometry:Linear Granulometry: lin_gran.pro granulo_linear.proGranulometry by Opening: granulo_open.pro granulo_open_general.pro

granulo_open_line.pro

* Indicates IDL subroutines



Global Thresholding. When the intensity level inthe ROI is known to have a threshold value (upper orlower), or known to fall in an interval, global thresh-olding is a powerful way of localizing it. The thresh-old limits and the interval boundaries can some-times be provided by the physical properties relatedto the gray-level/intensity.

For example, in x-ray radiography, tomography,ultrasound, and infrared imagery, the propertycould be intensity, x-ray attenuation coefficient,sound attenuation, or surface temperature, respec-tively. Very often however, the limits have to beinferred from the statistical distribution of the graylevels of pixels/voxels in the image. Specifically,these limits are determined from the “peaks” (ormodes) and “valleys” (separating two consecutivemodes) of the histogram of the distribution whenthey are prominent.

Global Clustering. Frequently, intensity of theimage pixels/voxels alone does not provide the desir-able separation between the ROIs, the background,and other objects. In such cases, one may want topursue this separation in a higher-dimensionalspace in a multivariate setting. In addition to theintensity, one might include other “features” of thepixels/voxels in the image, such as some statisticalparameters associated with a suitably defined neigh-borhood of them.

More generally, elements of the feature vectorcould represent gradient magnitude, gradient phase,intensities at multiple spectral bands, or otherattributes. It may also be some other measuredproperty of the pixel/voxel. The separation of theROIs from the background is then sought in themulti-dimensional feature space by an appropriate,unsupervised training procedure such as “cluster-ing.” Pixels/voxels that are “close” to one another inthe feature space as measured by their Euclidean orsome other distance, are considered “similar.” In aclustering procedure, these similar objects are clus-tered together. In this work we have used theK-means clustering algorithm.4,5

Constant False Alarm Rate (CFAR) Algorithm.When the intensities in an image vary systemati-cally, with a possible trend, the standard threshold-ing technique is unlikely to work. In such cases,detection can sometimes be implemented satisfacto-rily by means of the CFAR algorithm. This is particu-larly effective when the local variance of the inten-sity changes at a smaller rate than the local mean.

In this algorithm, one first subtracts the “local”mean from each pixel/voxel. Then a statistical test isapplied to each pixel/voxel to determine if it is differ-ent “significantly” from its background pixels/voxels.

Although computationally intensive, the algorithmallows for a pre-assigned false alarm rate in thedetection of ROIs.6

Region-Merging-Based Segmentation and3-D Connected Component Analysis. In thismethod, each 2-D slice is segmented by a region-growing/merging method with a label attached toeach segment by means of a connected compo-nent labeling algorithm. Adjacent connectedcomponents in consecutive slices are then puttogether and re-labeled to complete the 3-Dconnected component analysis.7,8

Segmentation by Watershed and 3-DConnected Component Analysis. This is the sameas above, except the 2-D segmentation is done bythe watershed algorithm.9

Identification

As indicated in an earlier paragraph, identifica-tion is frequently achieved by a “supervised learn-ing” procedure based on “ground truth” and isoutlined in the following steps. The “ground truth”here refers to the known classification categories ofthe training pixels/voxels.

Feature Computation. Various features arecomputed for “training” samples of known objecttypes. The nature of features varies widely from oneapplication to another. In this work we haveprovided routines to compute 2- and 3-D neighbor-hood statistics and extraction of first- and second-order statistics from these neighborhood statistics.

Feature Subset Selection. If the number offeatures is large, a smaller subset of these featurescontaining most, if not all, class discriminatoryinformation is often desirable. Frequently, due toresource constraints, the experimenter has a smallsample to work with, and as a result is not able toestimate accurately the distribution parameters in ahigher-dimensional feature space.

In a subset selection procedure, feature subsetsare selected based on their “separability measure,”which provides an index of how well the object typesin the feature space consisting of a given subset offeatures are separated.10

The simplest possible separability measure in amulti-dimensional feature space is based on theEuclidean distance between pairs of points repre-senting the selected features of samples from twopopulations. This applies when features may beassumed independent. The minimum of such pair-wise distances after appropriate normalizationgives the separability measure of the subset offeatures under consideration. When the features

FY 98 5-53



are correlated, one can replace the Euclideandistance by other distances, such as Bhattacharyaor Matusita.

Training and Classification. A suitably chosen“classifier” is then trained, based on the features ofthe training sample. A classifier is a system thatcreates class partitions in the feature space basedon labeled (known) samples. Based on the featurescomputed from a new sample, the trained classifierchecks which cell of the partition the new samplebelongs to and labels it accordingly. We have chosena non-parametric Bayes classifier for the presentwork. It is also termed the “Probabilistic NeuralNetwork” (PNN). The trained classifier is then usedto classify a sample as needed.11

Quantification

Having localized and identified the ROIs in theimage, the researcher would like to quantify theROIs in an appropriate manner. We have used thetechnique of morphological granulometry12,13 toimplement this capability. Traditional granulometryis an essential tool of quantitative morphologicalimage analysis. It consists of sequential applicationof openings (closings) with structuring elementsthat are monotonic.

The principle of granulometry can be expressedin simple terms as follows. Imagine a large collec-tion of “grains” of different sizes, filtered throughsieves of progressively increasing mesh sizesNumerical characteristics of the residues at eachstage are measured until nothing remains. Thesuccessive residues provide us with a quantitativemeasure of the grain size distribution.

In fact, by varying the shapes and orientationsof the mesh elements in the sieves, it is possibleto get a complete quantitative morphologicalcharacterization of the grains. The method wasfirst conceived by Matheron who called it“morphological granulometry.”

As a practical application, this method provides adirect computation of the size distribution of“grains”/“pores” in a binary image in one, two, orthree dimensions without having to do a connectedcomponent analysis first. An interesting feature inthis implementation is that one can detect the sizedistribution of the image grains along any specificdirection, or more generally, those fitting specificallyshaped objects with a given spatial orientation.

Applications

In this section we provide four typical applica-tions of the subroutines in our tool-kit. This willinclude 3-D images from computed tomographyusing x-ray and ultrasonic imagery.

1. Asphalt sample. The Federal HighwayCommission provided us with an asphalt sampleof approximate diameter 102 mm. The objectivewas to quantify such components as tar, aggre-gate, and porosity in the sample. A linear-arraydetector CT scanner (LCAT) operating at a maxi-mum tube potential of 160 keV (1.9 mA) wasused, with the data filtered intentionally to reducebeam-hardening to acquire a 3-D CT image. Theimage (Fig. 1) shows a cross-section measuring102 mm with a pixel size of 225 µm and a sliceplane thickness of 500 µm. Figure 2 shows theattenuation histogram of the slice.


4000Image histogram

Attenuation0.20.10 0.3

Freq

uen

cy

3000

2000

1000

0

Figure 1. Representative slice of the CT scan of an asphaltsample containing asphalt, rocks and air bubbles.

Figure 2. Histogram of the pixel attenuation in the CT imageshown in Figure 1.



The image is thresholded using the “valley”points v = [0.04, 0.2], resulting in the three“segments” shown in Figs. 3, 4, and 5. InFig. 3, attenuation less than 0.04, representsthe voids and background; attenuation between0.04 and 0.2, represents the asphalt; and thehigh attenuation voxels represent the inclusions.Pixels that are near zero represent air both insideand outside the sample. Thus this simple methodhas separated air from the rest of the sample. The first segment (Fig. 3a), shows that the

background (in white) and the air pockets(also in white) in the block have been sepa-rated. The second segment (Fig. 4b) shows therest, including a water droplet at the bottombut excluding the speckled bright spots evidentin Fig. 1. The third segment (Fig. 4c) showsthe speckles.

The histogram analysis (histo_analyze) andsegmentation (histo_seg) routines used in thisexample have been used on the 2-D slice butthey are equally applicable with similar resultsto the global 3-D data.

While thresholding separates some of theobjects, it fails when there are physically differ-ent objects with highly overlapping ranges ofintensities. This is the case with the asphalt,rocks, and water vial in this image. To addressthis problem we try to separate the objects in ahigher-dimensional “feature” space consisting ofa set of neighborhood statistics of each

pixel/voxel. Specifically, we choose a windowcentered at each pixel/voxel and compute a setof statistics for the pixels/voxels in the window.We then cluster the pixels/voxels (inclusionsexcepted) of the image into a fixed number, K, ofclusters, where K is the number of separableobjects known to be present in the image. Theresults of the clustering using K = 3 are shownFig. 4.

A tabulation of the percentages of pixelsrepresenting the four different objects within theaggregate is given in Table 2.

2. Aluminum part. A second example demon-strates the use of the CFAR 3-D code for isolat-ing regions suspected to be porous in 3-Dimagery obtained using a high-frequency ultra-sonic probe on an aluminum part. One of theobjectives in the porosity study is to get an esti-mate of the pore-size distribution. Figures 5ato d show a representative slice in the middlepart of the ultrasonic scan of the aluminumpart. Figure 5a shows the raw image. Figure5b shows the result of a pre-processing step,

FY 98 5-55

Table 2. Composition of the aggregate sample by percentage.

Rock 49Asphalt 29Void 16Inclusion 6

Figure 3. Binaryimage showing thethree segmentsobtained by usingthe threshold vector[0.04, 0.2] in theimage in Figure 1.

Figure 4. Threesegments showingthe asphalt, rocks,and the waterdroplet.



the absolute value of the intensity deviationfrom the mean image intensity. Figures 5c andd show the result of applying the CFAR 3-D codeto isolate the “anomalies” using a window sizeof 11 × 11 and 15 × 15, respectively, with thefalse alarm rate set at the five percent level.

3. A third example illustrates other routines forobject detection. We used a combination of theCFAR and a 3-D connected component analysisfor a 3-D segmentation. The data used are takenfrom the Active and Passive ComputedTomography (A&PCT) program for waste drumradioactivity assay. The objective here is tolocalize the regions of high radioactivity. Theframes in Fig. 6 represent 14 × 14 × 17 slicesof A&PCT data acquired at 414 keV. Slices 1 to15 arranged from top to bottom and left to rightin three rows are shown in Fig. 6.

The light blobs indicate the presence of highradioactivity. We applied the CFAR_3D algo-rithm to localize the radioactive objects in theimage. The result is shown in Fig. 7.

A 3-D connected component analysis led tothree distinct segments (not shown) consistingof 6, 16, and 44 voxels, respectively.

4. Our final example is taken from the porosity studyof a die-cast pre-stressed aluminum (A-356)tensile bar, to illustrate how methods of morpho-logical granulometry can be used to derive the1-, 2- and 3-D size distribution of pores (orgrains) using 3-D tomographic data. These inturn can be used to model metal compositebehavior under different kinds of stress.

Figure 8 shows a single slice of a 3-D CT scanof the aluminum sample using the PCAT systemwith 80 keV. The CT image has a pixel-resolution


(a) (b)

(c) (d)

Figure 5. (a) High-frequency ultrasoundc-scan image of analuminum part.Successive steps usedto isolate the poresinclude: (b) absolutedeviation from themean, a preprocess-ing step; (c) results ofthe application ofCFAR with a 3-Dkernel of 11 voxels to(b); and (d) results ofthe application ofCFAR with a 3-Dkernel of 15 voxels to(b).

Figure 6. Slices 1 to15 (top left is blank)of A&PCT dataacquired at 414 keV.



of 20.5 µm. It is of interest to find the size distrib-ution of the pores in the matrix along a chosendirection, in particular along the three axes.

One may also be interested in finding the 2-and 3-D size distribution of these pores. Thegranulometry routines developed in this workhave been used to find such distributionsapplied to the thresholded binary imageobtained from the original. Figures 9 and 10show the size distributions along the x- andz-axis, and Fig. 11 the same for the 3-Dvolume. If needed, a comparison of the two sizedistributions can be made at this stage.

Future Work

Three-dimensional image analysis resulting froma variety of probes such as x-ray CT, A&PCT, ultra-sonic scans, and NMR present a formidable chal-lenge to the data analyst. In this work we haveprovided a tool-kit based on the commercially avail-able software IDL to address some of the interestingclasses of problems that arise in such analyses.

Several possible extensions to improve andextend the capabilities of this tool-kit are beingplanned. First, like all useful tool-kits, there is theneed for a graphical user interface (GUI). The GUI

FY 98 5-57

Figure 7. The local-ized highly radioac-tive regions in slices 1to 15.

Figure 8. Representative CT slice of a pre-stressed A-356 die-cast aluminum tensile bar with pores showing in dark shade.

0.6

Pore size20100 30

Rel

. fre

que

ncy

0.4

0.2

0.0

Figure 9. Pore-size distribution along the x-axis in thetensile bar



allows the investigator to concentrate on the dataanalysis and inference process without gettingbogged down in the image-processing details.

Second, some of the algorithms used for granu-lometry in this tool-kit are of complexity O(n) orhigher, where n is the number of pixels/voxels in theinput image. Some recent developments in morpho-logical image processing on “fast” openings14,15 willallow us to replace these slow executing algorithmsby faster ones.

Third, we would extend single granulometry inthis work to “multivariate granulometry.” A singlegranulometry yields the size distribution of a singlevariable where structuring elements grow at thesame rate so that their relative sizes are constant.Multivariate granulometry16 allows each structuringelement to grow at its own chosen rate, independentof the scales of other structuring elements enablingbetter representation of the textural features. Atexture classification protocol using multivariategranulometry will capture the morphological charac-teristics of an image better and allow more accurateimage classification.

Acknowledgments

I am grateful to C. Logan, P. Roberson, D. Chinn,and J. Haskins for supplying the data shown in theapplications above. I am particularly grateful toH. Martz for carefully reviewing an earlier draft andsuggesting numerous improvements.

References

1. Rosenfeld, A., and A. Kak (1982), “Digital PictureProcessing,” 2nd ed., Vols. 1 and 2, Academic Press,Orlando, Florida.

2. Serra, J. (1982), “Image Analysis and MathematicalMorphology,” Academic Press, New York, New York.

3. IDL Version 5.0 (1997), Research Systems Inc.,Boulder, Colorado, March.

4. Ball, G. H., and D. J. Hall (1966), “ISODATA, A NovelMethod of Data Analysis and Pattern Classification,”International Communication Conference,Philadelphia, Pennsylvania, June.

5. Jain, A. K. (1989), “Fundamentals of DigitalImage Processing,” Prentice Hall, EnglewoodCliffs, New Jersey.

6. Reed, I. S., and X. Yu (1990), “Adaptive Multi-BandCFAR Detection of an Optical Pattern with UnknownSpectral Distribution,” IEEE Trans. on Acoustics,Speech and Signal Processing, June.

7. Levine, M. D. (1985), “Vision in Man and Machine,”McGraw-Hill, New York, New York.

8. Haralick, R., and L. Shapiro (1992), “Computer andRobot Vision,” Vol. 1, Addison-Wesley, Reading,Massachusetts.

9. Beucher, S., and F. Meyer (1993), The MorphologicalApproach to Segmentation: Watershed transforma-tion in “Mathematical Morphology in ImageProcessing,” E. Dougherty, ed., Marcel Dekker Inc.,New York, New York.


0.8

Pore size20100 30

Rel

. fre

que

ncy

0.6

0.4

0.2

0.0

1.0

Box side measure1050 15

Rel

. fre

que

ncy

0.8

0.4

0.6

0.2

0.0

Figure 10. Pore-size distribution along the z-axis in thetensile bar.

Figure 11. Pore-size distribution using openings by boxes.



10. Fukunaga, K. (1990), “Introduction to StatisticalPattern Recognition,” 2nd. ed., Academic Press, SanDiego, California.

11. Specht, D. F. (1990), “Probabilistic Neural Networks,”Neural Network, Vol. 3, No. 1, pp. 109–118.

12. Serra, J. (1982), “Image Analysis and MathematicalMorphology,” Academic Press, New York, New York.

13. Dougherty, E., J. Pelz, F. Sand, and A. Lent (1992),“Morphological Image Segmentation by LocalGranulometric Size Distributions,” Journal ofElectronic Imaging, Vol. 1, (1), January.

14. Crespo, J., R. W. Schafer, J. Serra, C. Gratin, and F.Meyer (1997), “The Flat Zone Approach: A GeneralLow-level Region Merging Segmentation Method,”Signal Processing, Vol. 62, pp. 37–60.

15. Vincent, L. (1994), “Fast Opening Functions andMorphological Granulometries,” SPIE, Vol. 2300,pp. 253–267.

16. Batman, S., and E. Dougherty (1997), “SizeDistributions for Multivariate MorphologicalGranulometries: Texture Classification and StatisticalProperties,” Optical Engineering, Vol. 36, (5), pp.1518–29.

FY 98 5-59


mage Recovery Techniques for X-Ray ComputedTomography in Limited-Data Environments


Introduction

Tomography is used throughout LLNL to investi-gate the composition and structure of objects nonin-vasively. Conventional CT algorithms have theadvantage that they are noniterative, hence, they arecomputationally efficient, but they suffer from a lackof flexibility in that they can’t incorporate priorinformation about the solution, nor can they useother than the squared error criterion to fit the data.

Because the image recovery problem associatedwith most tomography applications is ill-posed, whichis particularly true for limited-data situations, allpossible prior information about the unknown objectmust be used to produce high-quality images.Furthermore, although least-squares (that is, mini-mizing a sum of squares error function that measuresthe mismatch between the data and the model) isappropriate if the data has a Gaussian distributionwith known constant variance, using it when the datahas a non-Gaussian distribution can seriouslydegrade the quality of reconstructed images.

An example is counting problems (such as emis-sion tomography) where the data consists of particleor photon counts that typically have a Poisson distri-bution. Another example is outlier-corrupted data, inwhich the distribution function is not known exactly,but it is known that the data pixels are subject tooutliers that occur infrequently, but greatly distortthose pixels where they occur. These outliers resultfrom a variety of problems including a few bad detec-tors in a CCD array, improper assumptions about themodel, or “hits” from extraneous radiation or parti-cles. Outliers have highly deleterious effects on thereconstructed image when squared error is used.

The problems our techniques have been designedto handle can be described as follows. First there isa linear equation that models the relationship of theunknown to the data:

, (1)

where the vector represents the unknown imagewe wish to reconstruct; and A is a matrix that

x

ˆ ˆy Ax=

FY 98 5-61

There is an increasing requirement throughout Lawrence Livermore National Laboratory (LLNL)for nondestructive evaluation using x-ray computed tomography (CT). Conventional CT methods arenon-iterative algorithms that have the advantage of low computational effort, but are not sufficientlyadaptable to incorporate prior information or non-Gaussian statistics. In many cases, restrictions ondata acquisition time, imaging geometry, and budgets make it infeasible to acquire projection dataover enough views to achieve desired spatial resolution using conventional CT methods. Mostcurrently existing iterative tomography algorithms are based on methods that are time-consumingbecause they converge very slowly, if at all. The goal of our work was to develop a set of limited-dataCT reconstruction tools and demonstrate their usefulness by applying them to a variety of problems ofinterest to LLNL. In the second and final year of this project we continued our development of recon-struction tools and have demonstrated their effectiveness on several important problems.

Dennis M. Goodman and Jessie A. JacksonLaser Engineering DivisionElectronics Engineering

Maurice B. AufderheideDefense Systems

Erik M. JohanssonConsultantTracy, California

TA550 Goodman_qk 7/27/99 9:20 AM Page 5-61


represents such effects as geometry and absorptionthat lead to the expected data, , if were thetrue image.

The actual observed data is y, and the estimate ofthe true image is obtained by using an optimizationalgorithm to adjust so that is a best fit to theactual data according to some criterion function, say

, (2)

where the choice of the criterion function, as wenoted earlier, depends on the statistics of y.Consequently, our reconstruction problem involvesminimizing, with respect to , a function of theform

. (3)

This function is usually a negative log-likelihoodfunction, so the image we recover is the maximumlikelihood estimate of the unknown.1 Because ourproblems typically are ill-posed, we add penaltyparameters and impose prior constraints on x. Thisleads to the minimization problem:

. (4)

Here we are recovering a constrained penalizedmaximum likelihood estimate: the extra Euclideannorm and/or absolute value norm terms penalizelarge , and the parameters η and λ determine thedegree of penalty. It is also possible to include termsthat penalize derivatives of x, thereby imposing alarger penalty on its higher frequency components.

If outliers are present, we do not use the maxi-mum likelihood approach directly. Instead, weselect f(y, ) to make the reconstructed image lesssensitive to outliers. We give an example in thenext section.

The subset S represents upper and lower boundson the components of the image vector . Thesebounds provide crucial prior information about thesolution, and can greatly improve the quality of therecovered image. Because the unknown imageusually represents non-negative quantities such asabsorption or energy, the most common constraintis that all the components of be non-negative.Other constraints on subsets of pixels can resultfrom the known extent of the object and regions ofknown voids or occlusions.

The optimization problem we have describedabove is a difficult one for several reasons. The mostobvious is its very high dimensionality: both andy frequently have 106 or more components. Anotherreason is that bounds that we have imposed on the

x

x

x

y

x

x y x x x

x S = Argmin L ,

∈( ) + + η λ

1 2

L , ,y x y Axˆ ˆ( ) = ( )f

x

f y y, ˆ( )

y x

x ysolution require a constrained optimization tech-nique. Many iterative nonlinear minimization algo-rithms require storing an inverse of the matrix ofsecond partials of L(y, ) with respect to thecomponents of . This is clearly not possible forour problems because a 106-×-106 matrix is toolarge to store.

The answer for unconstrained problems is eitherthe conjugate gradient algorithm or the limitedmemory quasi-Newton algorithm. Both of these algo-rithms in effect store low-order approximations ofthe second-order information that is provided by theinverse matrix of second partials.

For bound-constrained problems, conventionaloptimization techniques2–4 usually allow only onevariable per iteration to attain a bound, so forvery large problems these techniques are veryslow because they spend too much time findingbounds. Consequently, standard iterative tomog-raphy algorithms5,6 are based on methods thatcan attain multiple bounds in an iteration.Unfortunately, however, they are essentiallysteepest descent techniques that use no second-order information about L(y, ) whatever, sothey converge very slowly, if at all.

Progress

Optimization Algorithm Development

Our ability to solve difficult tomography problemsis the result of two specialized optimization algo-rithms that we have developed. The first is an exten-sion of the conjugate gradient algorithm that incor-porates bound constraints on the variables.7 Thisconstrained conjugate gradient (CCG) algorithm isunique in that it incorporates a bending linesearchthat permits multiple bounds to be attained during asingle iteration. In previous years, the algorithm wasapplied with great success to a variety of practicalproblems.8–15

The second algorithm is a limited-memory quasi-Newton (QN) algorithm that permits upper andlower bounds on the variables. This year wecompleted the implementation of our QN algorithm,based on the derivations in Reference 16.

The very large dimensions of A typically makecalculating the matrix-vector product of Eq. 1 themost computationally expensive part of finding asolution. Which algorithm is most appropriatedepends on the structure of A.

In particular, the value of L(y, ) and a relateddirectional derivative must be calculated at eachnew sub-iteration within the bending linesearch.Completely recalculating Eq. 1 each time is very

x

x

x x




expensive, but in some cases a different approach ispossible. This is because it is necessary to recalcu-late only that part of Eq. 1 corresponding to compo-nents of that have either attained or left boundsduring a linesearch sub-iteration. For example, ifonly one component of has changed its boundstatus, then it is necessary to compute only onescalar-vector product, the product of that componentwith its corresponding column of A—rather than theentire vector-matrix product shown in Eq. 1. If A hasmany columns, this saves considerable effort.

If the columns of A are readily available, thenCCG appears to be the best choice. However, inmany tomography problems, Eq. 1 is not computedin the usual manner. Often A is a discretized versionof the forward projection operator; therefore, itrepresents a set of line integrals, and it is verysparse. Consequently, Eq. 1 is computed via a rulethat calculates only the line integrals and wastes notime on the large portions of A that consist of zeros.

Another example is where A is the kernel of ashift-invariant blurring function. In this case, using aconvolution algorithm based on the fast Fouriertransform (FFT) is by far the most efficient methodfor computing Eq. 1.

In some of our future work on the AdvancedHydrotest Facility (AHF)17 we anticipate that A willbe of the form A = BC, where C is a projectionmatrix and B is a blur kernel that represents theeffects of detector and source geometries. In thiscase, both a sparse matrix rule and an FFT convolu-tion algorithm are required. In all the casesdiscussed in this section, the programming effortrequired to calculate individual columns of A isconsiderable, and often not worthwhile.

The advantage of the QN method is that bendingis used first on a low-order quadratic approximationto L(y, ) to produce a direction for the linesearch.This greatly reduces the need for bending during thelinesearch itself, yet it still permits the algorithm toattain multiple bounds during an iteration. In the QNalgorithm we have included the options both toperform additional bending in the linesearch if thecolumns of A are available, and to perform no bend-ing in the linesearch if they are not.

Robust Tomography Algorithms

This year we completed work on and demon-strated the usefulness of our robust tomographytechnique. The squared error function is

, (5)

f rii

N

y y r r, ˆ ( ) = ==∑ 2

1

'

x

x

x

where the residual vector r = y – is one indicationof mismatch between the observed data and themodel of the data.

If Eq. 5 is the criterion function in Eqs. 2 and3, then the reconstructed image defined in Eq. 4will be very sensitive to outliers. For example,suppose N = 10,000 in Eq. 5, and the magnitude ofa typical component of the residual is 1.0, exceptfor one component, rk, whose value is 100.0. Thenrk contributes roughly as much to the squarederror as do all the other components combined,and the minimization algorithm will try very hardto match the outlier, yk, at the expense of all theother data points.

The solution is to use a robust criterion func-tion18,19 that reduces the influence of larger residu-als. The most common robust criterion is

, (6a)

where

. (6b)

The function g(r) transitions smoothly from asquared penalty to a linear penalty, thereby reducingthe influence of large residuals. The choice of c isdata-dependent.18,19

We have demonstrated both that CCG is veryeffective in minimizing Eq. 4 when Eq. 6 is thecriterion function, and that this approach greatlyimproves image quality in the face of outliers. Wepresent a simulated example. The true unknown isshown in Fig. 1, and the resulting projection datawith noise added is shown in Fig 2.

This data is in the form of a sinogram, a 2-Dimage created by displaying all the ray sums at oneangle vs all angles (projections) obtained. The noiseis a Gaussian mixture, with probability 0.99 that thenoise at a data pixel is Gaussian with standard devi-ation σ, but with probability 0.01 it is Gaussian withstandard deviation 100σ. Consequently, there is anoutlier 1% of the time.

These outliers are obvious as speckles in Fig. 2.The reconstructed image using Eq. 5 is shown inFig. 3; the reconstructed image using Eq. 6 isshown in Fig. 4. Outliers caused the streaks inFig. 3. Our robust technique reduced the rms errorbetween reconstruction and true object by a factorof two. Although robust statistical methods have beenapplied to a variety of problems, to our knowledgethis is their first application to tomography. Theeffectiveness of CCG made this possible.

g rr if r c

c r c if r c( ) =

≤

≤

–

2

22

f g rii

N

y y, ˆ ( ) = ( )=∑

1

y

FY 98 5-63




0

0.01

0.02

0.03

0.04

0.05

0.06

0.07

0.08

0.09

0.10

20 40 60 80 100 120

20

40

60

80

100

120

Figure 1. Originalobject used for CCGsimulated example.

20

40

60

80

100

120

140

160

180

200

20 40 60 80 100 120

10

5

0

-5

-10

Figure 2. Sinogram ofprojections of objectshown in Figure 1,with Gaussianmixture noise added.



FY 98 5-65

20

40

60

80

100

120

20 40 60 80 100 120

0.12

0.10

0.08

0.06

0.04

0.02

0

20

40

60

80

100

120

20 40 60 80 100 120

0.10

0.08

0.06

0.04

0.02

0

Figure 3.Reconstruction fromprojection data inFigure 2, usingsquared error crite-rion.

Figure 4.Reconstruction fromprojection data inFigure 2, usingrobust technique.



We have also successfully tried other robustcriteria. Although we presented only a simulatedexample, many tomography problems are plaguedby outliers, and this method has promise for avariety of real problems.

Waste Drum Assay

Using emission tomography to characterize mixedwaste drums has been studied at LLNL for severalyears.20 An active and passive computed tomogra-phy (A&PCT) technique has been developed thatfirst uses an external radioactive source and activetomography to map the attenuation within a wastebarrel. This attenuation map is used to define thematrix A for the passive or emission tomographyproblem that is of interest.

The emission tomography part of the waste drumproblem gave us opportunities to demonstrate theeffectiveness of our techniques and to develop themfor general emission tomography problems.

At each detector position we acquire the entiregamma-ray spectrum, and two counts of gamma-rayemissions are taken. The first count is in the regionof a spectral peak of the isotope of interest; thesecond in a region just outside this spectral peak.The purpose of the second measurement is to deter-mine the level of background radiation and removeits effects on the first measurement.

In previously developed algorithms, the netcounts due to the isotope were obtained by subtract-ing the second measurement from the first. Twomaximum likelihood expectation-maximization(MLEM) algorithms were developed, UCSF-MLEM21

and APCT-MLEM,22 and applied to the correcteddata to obtain a 3-D image of isotope activity. Thesum of counts over all the image voxels is related toan estimate of isotope activity within the drum.Unfortunately, subtracting the two counts and thenusing a maximum likelihood algorithm on the netcounts is not a correct application of the likelihoodprinciple. Furthermore, this approach can violatephysical reality because there is a non-zero proba-bility that a net count will be negative.

The two MLEM algorithms lack the flexibility toimplement the correct log-likelihood function.Last year we derived the correct log-likelihoodfunction;23 this year we developed a new algo-rithm, APCT-CCG, for the waste drum problem,and we studied its behavior on both real andsimulated data.24

The flexibility of CCG made implementing the log-likelihood function relatively simple. It was neces-sary only to select the appropriate criterion functionthat accounted for Poisson statistics and incorpo-rated peak and background measurements at eachdetector location. Figure 5 compares applying thethree algorithms to real data from known Pu-239


-50.00

-40.00

-30.00

-20.00

-10.00

0.00

10.00

20.00

30.00

40.00

50.00

Perc

ent

dif

fere

nce

fro

m a

ctua

l dat

a

Ba-

alp

ipe

Ba-

ban

a

Ba-

bas

and

Ba-

ph

anto

m

Pu

-c1

5g

20

Pu

-o1

5g

20

Pu

-pd

p_5

.c2

Pu

-pd

p_1

7.c

2

Pu

-pd

p_5

0.c

2

Pu

-pd

p_8

0.c

2

Pu

-pd

p_1

60

.c2

Pu

-pd

p0

03

.c3

Pu

-pd

p0

04

.c3

Pu

-pd

p3

.6g

Data sets

UCSF-MLEMAPCT-MLEMAPCT-CCG

Figure 5. Comparison of assay accuracy for three reconstruction methods.



sources. APCT-CCG was out-performed on only oneof the seven sources. Although the voxel sum is themost important parameter for waste drum assay,the reconstructed image is more important for mostemission tomography problems. APCT-CCG also dida much better job of reconstructing the image.

A test case was generated using simulated data.A simulated 3-D image was created with threeslices. Each slice is 14 × 14 voxels. A point radioac-tive emission source of 30,350 counts was placed onthe center slice at voxel location (5, 5), that is, justoff the center of the slice. Using the system matrix,the image was forward projected to create threesinograms. A level background equal to the maxi-mum signal strength was added to these sinograms.The level was carefully chosen to be consistent withand representative of empirical data.

The simulated sinograms were randomized bypassing them through a Poisson random generator.

Another set of sinograms were created with thesame background level. These background sino-grams were also randomized. Both the gross andbackground sinograms were used as input for theAPCT-MLEM and APCT-CCG codes. The results areshown in Fig. 6.

The APCT-MLEM image is spread out over 3 × 3voxels within each slice and across all three slices.Its total assay yields 36,710 counts, that is, 121% ofthe actual value. The APCT-CCG results are morelocalized and its assay value of 9,936 counts ismuch closer to the original 30,350 counts, that is,within 1%.

The other important observation is that the APCT-CCG-code-calculated sinograms are more represen-tative of the original source data than the sourceplus background, which is not the case of the APCT-MLEM sinogram results. We also found this to bethe case for real data.20

FY 98 5-67

Simulation No backgroundAverage Net

= 2 counts

True image30350 counts

APCT-MLEMReconstruction36710 countsR = 121%

APCT-CCGReconstruction30072 countsR = 99%

Simulation withaverage background

= 20 counts

APCT-MLEMCalculated

APCT-CCGCalculated

Sinograms Tomographic images

0 2 4 6 8 10 12

1050 15 20 25

x 1.00e+03

Figure 6. Simulated image and sinogram reconstruction.



For this problem it became obvious to us thatmodel selection and validation are important issues,particularly when accurate estimates of parameters(for example, isotope activity) are as critical as goodimages. In the case of Gaussian noise problemswhere the squared error function is used, tomogra-phy can be viewed as a linear regression problem, sothe usual χ2 tests can be applied to the squared errorto measure goodness of fit and to choose betweenmodels. The sinogram of the residuals is a goodimage to observe the adequacy of a particular model.

However, this is not the case if the noise is notGaussian. In studying the waste drum problem, werealized that statistically it can be viewed as ageneralized linear model in which there is a linearrelationship between the unknown and the data, butthe statistics are not necessarily Gaussian.19,25,26

This is basically the model we defined in theIntroduction. In this context it is possible to slightlymodify L(y, ) so that it becomes a deviance func-tion that has the same minimum with respect to ,but exhibits behavior that is approximately χ2.

Similarly, it is possible to define other sinogramimages that behave as the usual residual image doesfor the squared error case. We have applied theseideas to the waste drum problem, and they will beuseful for other tomography problems as well.

Neutron Imaging

LLNL is currently developing a high-energy (10 to15 MeV) neutron imaging system for use as an NDE

xx

tool in support of the Enhanced SurveillanceProgram (ESP). This approach to tomographypromises to be a powerful technique for probing theinternal structure of thicker objects that may beopaque to x rays and lower energy neutrons.

Imaging experiments using neutron radiographywere conducted at the Ohio University AcceleratorLaboratory (OUAL) in FY-98. The object imaged wasa right-circular Pb cylinder with an outer diameterof 4 in. and a 2-in.-diameter polyethylene insert, asillustrated in Fig. 7. The insert was split into twohalf-cylinders with one serving as “blank” and theother having a series of 10-, 8-, 6-, 4-, and 2-mm-diameter holes machined to depths of 0.5 in. into itsouter (curved) surface. The areal density of theassembly ranged from 62.38 g/cm2 (along thecenterline) to 99.9 g/cm2 (along the limb of the poly-ethylene insert).

Reconstructions of this object using both filteredbackprojection and our CCG algorithm with non-negativity constraints are shown in Fig. 8. The supe-riority of the CCG reconstruction is evident.Although we used the squared error criterion withCCG, in fact all the raw data acquired during theOUAL experiments initially bore random sharpspikes rising several hundred to several thousandcounts above the local average.

These spikes were due primarily to cosmic raystrikes in the CCD detector used to collect the data.Currently these spikes are removed by preprocessingprior to applying the tomography reconstructionalgorithms. However, such data is a perfect candidate


4.00 in. Pb poly Pb poly

2.00 in.Top view

Joint between half cylinders

4.00 in.

0.50 in.Figure 7. Neutronimaging object.



for the robust techniques we developed this year,and we plan to apply them to neutron data in thenear future. For a detailed description of the neutronimaging experiments, see Reference 27.

The Advanced Hydrotest Facility

Last year we adapted our CCG algorithm to conebeam tomography problems. This year we used theresulting algorithm, CCG_Cone, extensively, to studylimited-view reconstruction for LLNL‘s AHF. This codehas been an indispensable tool in these studies. It hasbeen used to study the efficacy of reconstruction as afunction of number of views, as well as how the orien-tation of views affects the quality of the reconstruction.

These studies have allowed the AHF design groupto make recommendations on how many views areneeded, as well as where they should be placed. Inaddition, CCG_Cone has been used to study the effectof constraints on the quality of reconstructions. Ithas been found that a judicious use of constraintswill help the AHF to achieve its objectives. Moreprogrammatic work will continue in this area.

Pulsed Photothermal Radiography

In collaboration with researchers at other institu-tions, we continued to apply our optimization algo-rithms to the problem of pulsed photothermal radi-ography (PPTR). This is essentially a tomographic

FY 98 5-69

-0.020

-0.010

-0.000

0.010

0.020

Rec

on

stru

cted

in

ten

sity

lev

el

0 200 400 600 800 1000

Lineout through image center(4-cm thick scintillator, 50 mm f/1.0 lens)

FBP reconstructionCCG reconstruction

Image column

(c)

(a) (b)

Constrained Conjugate Gradient (CCG) Standard Filtered Back-Projection (FBP)

Figure 8.Reconstruction ofobject from neutronimaging data, usingCCG (a) and filteredback-projection (b).



method that inverts data from the time evolution ofthe heat equation, rather than from x-ray projectiondata, to see inside an opaque object. This year weobtained the first high-quality 3-D images of port-wine stain blood vessels,28,29 and we demonstratedthe feasiblity of parallelizing our CCG algorithm.30

Future Work

In future work we plan to continue refining ouralgorithms and applying them to practical problems.Immediate plans include parallelizing the algo-rithms, particularly the implementations of Eq. 1;further refinements of the forward projection model,including implementing the AHF blurring model; andfurther investigations into applying the generalizedlinear model formalism to tomography.

References

1. Poor, H. V. (1994), An Introduction to SignalDetection and Estimation, 2nd ed., Springer-Verlag.

2. Gill, P. E., W. Murray, and M. H. Wright (1981),Practical Optimization, Academic Press, New York,New York.

3. Fletcher, R. (1987), Practical Methods ofOptimization, 2nd ed., John C. Wiley and Sons, NewYork, New York.

4. Luenberger, D. G. (1984), Linear and NonlinearProgramming, 2nd ed., Addison-Wesley, Reading,Massachusetts, pp. 214–220.

5. Stark, H., ed. (1987), Image Recovery: Theory andApplication, Academic Press, New York, New York.

6. Azevedo, S. G. (1991), Model-Based ComputedTomography for Nondestructive Evaluation, PhDThesis, Lawrence Livermore National Laboratory,Livermore, California (UCRL-LR-106884), March.

7. Goodman, D. M., E. M. Johansson, and T. W.Lawrence (1993), “On Applying the ConjugateGradient Algorithm to Image Processing Problems,”Chapter 11 in Multivariate Analysis: FutureDirections, C. R. Rao, ed., Elsevier SciencePublishers.

8. Goodman, D. M., T. W. Lawrence, E. M. Johansson,and J. P. Fitch (1993), “Bispectral SpeckleInterferometry to Reconstruct Extended Objects fromTurbulence-Degraded Telescope Images,” Chapter 13in Handbook of Statistics, Vol. 10: Signal Processingand its Applications, N. K. Bose and C. R. Rao, eds.,North Holland, Amsterdam.

9. Kolman, J., W. S. Haddad, D. M. Goodman, and K. A.Nugent (1994), “Application of a ConstrainedOptimization Algorithm to Limited View Tomography,”Proc. SPIE Conf., San Diego, California, July.

10. Haddad, W. S., J. E. Trebes, D. M. Goodman, H. R.Lee, I. McNulty, E. H. Anderson, and A. O. Zalensky(1995), “Ultra High Resolution Soft X-RayTomography,” Proc. SPIE Conf., San Diego, California,July.

11. Somoza, J. R., H. Szöke, D. M. Goodman, P. Beran, D.Truckses, S.-H. Kim, and A. Szöke (1995),“Holographic Methods in X-ray Crystallography IV. AFast Algorithm and its Application to MacromolecularCrystallography,” Acta Crystallographica, Vol. A51.

12. Milner, T. E., D. M. Goodman, B. S. Tanenbaum, andJ. S. Nelson (1995), “Depth Profiling of Laser-HeatedChromophores in Biological Tissues by PulsedPhotothermal Radiometry,” Journal of the OpticalSociety of America-A, Vol. 12 No. 7, July.

13. Milner, T. E., D. M. Goodman, B. S. Tanenbaum, B.Anvari, L. O. Svasand, and J. S. Nelson (1996),“Imaging of Laser Heated Subsurface Chromophoresin Biological Materials: Determination of LateralPhysical Dimensions,” Physics in Medicine andBiology, Vol. 41.

14. Milner, T. E., D. J. Smithies, D. M. Goodman, A. Lau,and J. S. Nelson (1996), “Depth Determination ofChromophores in Human Skin by PulsedPhotothermal Radiometry,” Applied Optics, June.

15. van Gemert, M. J. C., J. S. Nelson, T. E. Milner, D. J.Smithies, W. Verkruysse, J. F. de Boer, G. W.Lucassen, D. M. Goodman, B. S. Tanenbaum, L. T.Norvang, and L. O. Svaasand (1997), “Non-InvasiveDetermination of Port Wine Stain Anatomy andPhysiology for Optimal Laser Treatment Strategies,”Physics in Medicine and Biology, Vol. 42.

16. Byrd, R. H., J. Nocedal, and R. B. Schnabel (1994),“Representations of Quasi-Newton Matrices and TheirUse in Limited Memory Methods,” Math. Prog., Vol.63, No. 2, January.

17. “Advanced Hydrodynamic Radiography TechnologyDevelopment Plan” (1996), Predecisional Draft, USDepartment of Energy, Defense Programs, Office ofResearch and Inertial Fusion, February 16.

18. Huber, P. J. (1981), Robust Statistics, John C. Wileyand Sons, New York, New York.

19. Myers, R. H. (1990), Classical and ModernRegression with Applications, PWS-Kent, Boston,Massachusetts, Chapter 7.

20. Martz, H. E., G. P. Roberson, D. C. Camp, D. J.Decman, J. A. Jackson, and G. K. Becker (1998),“Active and Passive Computed Tomography MixedWaste Area Final Report,” Lawrence LivermoreNational Laboratory, Livermore, California,November.




21. Brown, J. K., K. Kalki, J. A. Heanue, and B. H.Hasegawa (1995), “Quantitative SPECTReconstruction Using Multiray ProjectionIntegrators,” Conference Record 1995 IEEE NuclearScience Symposium and Medical Imaging Conference,Vol. 2, pp. 1272–1276.

22. Keto, E., S. G. Azevedo, G. P. Roberson, D. J.Decman, H. E. Martz, and E. M. Johansson (1995),“Spatial Resolution Versus Signal to Noise inQuantitative Tomography,” Proceedings of the 4thNondestructive Assay and NondestructiveExamination Waste Characterization Conference, SaltLake City, Utah, October 24–26, pp. 405–420.

23. Goodman, D. M. (1997), “Maximum LikelihoodEstimation with Poisson (Counting) Statistics forWaste Drum Inspection,” Lawrence LivermoreNational Laboratory, Livermore, California (UCRL-ID-127361), May.

24. Jackson, J. A., D. M. Goodman, G. P. Roberson, andH. E. Martz (1998), “An Active and Passive ComputedTomography Algorithm with a Constrained ConjugateGradient Solution,” 6th Nondestructive Assay WasteCharacterization Conference, Salt Lake City, Utah,November 17–19.

25. McCullagh, P., and J. A. Nelder (1989), GeneralizedLinear Models, 2nd ed., Chapman and Hall, London.

26. Dobson, A. J. (1990), An Introduction to GeneralizedLinear Models, Chapman and Hall, London.

27. Hall, J., F. Dietrich, C. Logan, and G. Schmid (1998),“Development of High-Energy Neutron Imaging inSupport of Enhanced Surveillance ProgramApplications,” Lawrence Livermore NationalLaboratory, Livermore, California, in preparation.

28. Shoari, S., N. Bagherzadeh, D. Goodman, T. E. Milner,D. J. Smithies, and J. S. Nelson (1998), “A ParallelAlgorithm for Pulsed Laser Infrared Tomography,”Pattern Recognition Letters, (19)5-6, pp. 521–526,April.

29. Milner, T. E., S. A. Telenkov, B. S. Tanenbaum, J. S.Nelson, and D. M. Goodman (1998), “Non-invasiveEvaluation of Biological Materials Using PulsedPhotothermal Tomography,” Proc. Winter AnnualMeeting of ASME, Anaheim, California, November.

30. Telenkov, S. A., D. M. Goodman, B. S. Tanenbaum, J.S. Nelson, and T. E. Milner (1998), “Infrared Imagingof laser-heated port wine stains,” Annual Meeting ofthe Optical Society of America, Baltimore, Maryland,October.

FY 98 5-71


Supporting Technologies

Robert T. Langland

material models were done for the aluminum alloy6061-T6 and the titanium alloy Ti-6Al-4V. Theseplasticity and failure models apply in tension,compression and shear, and are used in the NIKEand DYNA family of computer codes.

The third project, “Uniform Etching of 85-Cm-Diameter Grating,” has developed an ion etchingprocess to build 85-cm-diameter optical gratingsystems of silicon.

The fourth project, “Distributed Sensor InertialMeasurement Unit,” has been funded to refine exist-ing theory to develop, build, test and eventually usean accelerometer-based inertial measurement unit(IMU). This project uses six very sensitiveaccelerometers with special electronic signal condi-tioning. These accelerometers are arranged in a crit-ical configuration that permits the definition of thefull six degrees of freedom demanded by the applica-tion. This approach represents a departure from theuse of laser-based rate gyro IMUs, which could notbe used for our application.

The fifth project, “Fiber-Based Phase-ShiftingDiffraction Interferometer for Measurement andCalibration of the Lick Adaptive Optics System,” istransforming a new spherical wavefront-basedphase-shifting interferometer into a practical opti-cal measurement system. This project and the oneon etching represent examples of an underlyingand broad expertise in the design and fabricationof optical systems.

All the projects are working at extreme limits ofspace and/or time. We are attempting to design,fabricate and perform in areas that are pushing thetechnologies beyond their current limits. In oursupporting technologies, we are exploring and creat-ing new boundaries to meet LLNL’s programmaticneeds and goals for the future.

The five centers of excellence in the EngineeringDirectorate at Lawrence Livermore NationalLaboratory (LLNL) are made up of enabling tech-nologies that are essential for these centers to beworld class.

However, in addition to the enabling technologiesin each center, the Engineering Directorate also hasa broad set of supporting technologies that make upthe complete capabilities of engineering at LLNL.These allow engineering projects to accomplishspecific tasks and make it possible for a program tomeet one or more of its objectives or goals.

When appropriate, the Engineering Directoratesponsors work in these supporting technologies.Many efforts are small and do not require significantfunding; however, there are some that demandsignificant resources. The five articles in this sectionrepresent some of these more substantial efforts,which are very broad and diverse.

First, we have sponsored a project in multi-scalematerial modeling that complements a StrategicInitiative funded by LLNL’s Laboratory DirectedResearch and Development Program. This project,“Modeling of Anistropic Inelastic Behavior,” is refin-ing the theory of finite plasticity. It is coordinatedwith research funded by the National ScienceFoundation at the University of California atBerkeley. A Ph.D. candidate is using LLNL facilitiesto develop and carry out very sophisticated testingto map yield surfaces of real materials. The resultsof this testing will eventually be used in computercodes that have been developed through the leader-ship of LLNL’s Computational Engineering Center.

The second research project, “Modeling Large-Strain, High-Rate Deformation in Metals,” has usedHopkinson’s bar testing to develop a new model forimportant materials over a very large range of strainrates, from 104 s–1 to 10–4 s–1. The testing and



6


Contents

6. Supporting Technologies

OverviewRobert T. Langland

Modeling of Anistropic Inelastic BehaviorDaniel J. Nikkel, Jr., Arthur A. Brown, and James Casey ...........................................................................6-1

Modeling Large-Strain, High-Rate Deformation in MetalsDonald R. Lesuer, Mary M. LeBlanc, and Gregory J. Kay ...........................................................................6-7

Uniform Etching of 85-Cm-Diameter GratingSteven R. Bryan, Jr. and David L. Sanders...............................................................................................6-17

Distributed Sensor Inertial Measurement UnitCarlos A. Avalle and John I. Castor..........................................................................................................6-23

Fiber-Based Phase-Shifting Diffraction Interferometer for Measurement and Calibration of the Lick Adaptive Optics SystemEugene W. Campbell and Jong R. An .......................................................................................................6-27


odeling of Anisotropic Inelastic Behavior


Introduction

The ability of numerical simulations to predict thebehavior of systems involving materials undergoinglarge deformations is contingent upon having a real-istic model of the behavior of the materials involved.Such models must be accurate in the full range ofpossible loading conditions that the materials maybe subjected to. Use of overly simplified models inregimes where they are not well suited can seriouslycompromise the validity of a simulation. Many prob-lems of engineering interest involve metal undergo-ing large deformation under multiaxial states ofstress. The need for reliable models for these appli-cations can hardly be overemphasized. As will beseen, simple models for plasticity commonly used innumerical codes do not accurately predict materialbehavior under these conditions.

From the macroscopic perspective, polycrys-talline metals subjected to loads or deformationsinitially exhibit elastic (reversible) behavior. Thematerial response is path-independent and there is aone-to-one correspondence between stress andstrain. However, if the deformation or loads becomesufficiently large, the material begins to exhibit plas-tic behavior (that is, there is no longer a one-to-onecorrespondence between stress and strain, the

response is dependent on the loading path taken toreach a given state of deformation, and residual—plastic—deformations remain after external loadsare removed). This gives rise to the theoretical ideal-ization of an elastic-plastic material, and in particu-lar, to the notion of a yield function1 denoted by

(1)

This function, a key ingredient of the constitutivetheory of elastic-plastic materials, describes theboundary between stresses (or strains) that result inonly elastic behavior, and those that result in inelastic deformation (Fig. 1). In Eq. 1, skl denotesthe components of the stress tensor; ekl denotes thecomponents of the strain tensor; denotes thecomponents of the plastic strain tensor; κ is a scalarmeasure of work hardening; and the ellipses repre-sent other inelastic state variables that may bepresent, depending on the constitutive theory.

Annealed polycrystalline metals typically exhibitisotropic behavior with respect to a reference config-uration; that is, at a given point in the material, thematerial response of a specimen carried out in anydirection is the same. This includes the elasticbehavior and the initial yield behavior. However,significant processing of materials, or large plastic

eklp

f s e g e ekl klp

kl klp, , , , , , .κ κK K

=

FY 98 6-1

We are working to develop better constitutive equations for polycrystalline metals. An experimen-tal capability, developed at Lawrence Livermore National Laboratory (LLNL), is being used to studythe yield behavior of elastic-plastic materials. We are directly determining the multi-dimensional yieldsurface of the material, both in its initial state and as it evolves during large inelastic deformations.These experiments provide a more complete picture of material behavior than can be obtained fromtraditional uniaxial tests. Experimental results show that actual material response can differ signifi-cantly from that predicted by simple idealized models. The yield surface, and its mathematical repre-sentation, is an essential component of the constitutive theory for nonlinear anisotropic elastic-plasticmaterials, and is the main focus of the present project.

Daniel J. Nikkel, Jr.New Technologies Engineering DivisionMechanical Engineering

Arthur A. Brown and James CaseyUniversity of CaliforniaBerkeley, California

605 Nikkel_qk 7/27/99 9:25 AM Page 6-1


deformations, can result in behavior which isanisotropic, where material response in differentdirections is quite different. Examining themicrostructural processes which give rise to inelas-tic behavior, isotropic behavior can be seen as beingdue to the random orientation of grains of material,each of which has particular orientations and prop-erties. As a consequence of some types of process-ing, or due to large inelastic deformations, theinitially random grain orientations can becomealigned, resulting in anisotropic behavior.

For fixed values of the inelastic variables, theyield condition described by f = 0 (or g = 0) can beinterpreted geometrically from the point of view ofstress space (or strain space), the multi-dimensionalspace whose axes are the components of stress (orstrain), as a surface that bounds the region in whichonly elastic behavior occurs (the elastic region).

As long as the loading of the material is suchthat the current state is enclosed by the yieldsurface, the material responds elastically. But, ifthe loading path intersects the yield surface andtries to cross it, inelastic behavior occurs andplastic deformation results. The current statenever moves outside the yield surface, but insteadthe surface is carried along with it. Typically theyield surface changes shape as the inelasticdeformation increases. In addition to the yieldfunction, the constitutive theory includes evolu-tion equations for the inelastic variables duringloading (g = 0, > 0), such as that for the plas-tic strain:

,˙ , , , ˆe e e gklp

kl mn mnp=

ρ κ K

g

.(2)

Here ρkl is a constitutive response functionwhich is independent of the rates of stress orstrain. For a broad class of materials, under aphysically reasonable assumption regarding workin closed cycles in strain space, ρkl can bereplaced with the product of scalar function andthe normal to the yield surface in stress space,thus requiring the specification of only one addi-tional scalar response function.1 For specialclasses of materials, this scalar function is deter-mined from the yield function and hardening anddoes not require an independent specification.

Most models for plasticity of metals implementedinto numerical codes use a yield criterion that corre-sponds to a fixed shape of the yield surface (forexample, elliptical in the case of the Mises yieldcriterion). What distinguishes different models ishow the yield surface is assumed to evolve. Forexample, it may translate rigidly, or alternativelychange its size while maintaining its shape, or followsome combination of these simple hardening laws.

While the initial yield surface of isotropic mate-rials may be represented reasonably well by anellipse, subsequent to even moderate plasticdeformation, the shape of the yield surface in realmaterials can change significantly (Fig. 1). Forthis reason, simple representations of the yieldfunction will be satisfactory only under veryrestrictive loading conditions (for example, monot-onic or uniaxial), and are totally inadequate forgeneral multiaxial loading conditions where loadscan reverse and change direction during thehistory of loading.

ˆ ˙gg

ee

mnmn≡ ∂


Shea

rin

g s

tres

s (p

si)

400020000-2000-4000-6000-3000

-2000

-1000

0

1000 C2

C1 A

1

A2

B2

2000

3000

0

Initial

Axial stress (psi)

Figure 1. Measuredpoints on yieldsurfaces in 2-D stressspace from threespecimens of 1100aluminum. Thesubsequent yieldsurfaces show significant deviationfrom an idealizedellipse, even thoughthe strains involvedare moderately small.

605 Nikkel_qk 7/27/99 9:26 AM Page 6-2


In view of these considerations, and motivated bythe fact that the vast majority of experimental dataon polycrystalline metals that is available is foruniaxial (and generally monotonic) loading, wedeveloped an experimental capability to map out theyield surface at various fixed states of large inelasticdeformation under multiaxial states of loading. Bydetermining the yield surfaces on a single specimenat multiple fixed states, the evolution of the yieldsurface during plastic deformation can be observed.This data provides the basis for developing improvedconstitutive equations for polycrystalline metals.

Progress

This project is a combination of a program ofnovel experiments characterizing inelastic materialbehavior, together with an effort to develop bettermaterial models for implementation into numericalanalysis codes. This year, the primary effort hasbeen on the experimental component of theproject. Work has also begun examining issuesrelated to numerical implementation of anisotropicplasticity models.

Experiments

The first part of this project involves experi-ments to directly measure the yield surface. Thethin-walled tension-torsion specimen designed foruse in a multiaxial MTS hydraulic testing machineis shown in Fig. 2. The experimental determinationof the yield surface of the material is carried out byloading a specimen under multiaxial conditions andprobing until the point of yield is reached, thenbacking off and probing in a different direction instress space (and in strain space) until the nextyield point is found. This process is repeated untilthe entire surface is mapped out. The sensitivenature of the measurements being made requirescareful attention to the issues of specimen designand preparation, experimental methodology, andinterpretation of the data.

The general description of the experiments andthe difficulty in carrying out these measurementshas been discussed previously.2 The present discus-sion will focus on refinements that have been madeduring FY-98.

The surface that we are trying to map representsthe yield surface at an arbitrary fixed inelastic state.Ideally, all points on a given yield surface should bedetermined without inducing any further plastic defor-mation to the specimen. In practice, however, a pointon the yield surface can only be determined by reach-ing, and slightly exceeding, the yield point. Each time

the yield point is exceeded in this way, the inelasticstate, and the yield surface itself, are slightly changed.

For our purposes, it is important to minimize thisdistortion of the yield surface. To characterize agiven surface, a number of points on it must belocated (probably a minimum of 10) while changingthe inelastic state (hence, the surface itself) as littleas possible. The way in which yield is defined experi-mentally can significantly affect the yield surfacewhich is determined.

A number of alternative definitions have beeninvestigated, and the effects of different definitionson the resulting measured yield surfaces have beenstudied. The procedure which has been developedcan detect yield without producing a plastic strainmuch greater than 5 x 10–6. We refined the experi-mental procedure to reduce a number of sources oferror and have demonstrated that the methodologyfor determining yield surfaces is repeatable.

The importance of rate effects has also beeninvestigated. Even in materials which are notconsidered highly rate-sensitive, due to the

FY 98 6-3

4 in.

Figure 2. Thin-walled biaxial test specimen that can besubjected to tension, compression and torsion, used tomeasure points on the yield surface of the material. Theyield surface at several different states can be measuredfrom one specimen.

605 Nikkel_qk 7/27/99 9:26 AM Page 6-3


desired accuracy in measuring the yield surface,it was found necessary to run the tests veryslowly, increasing the time required to completeeach measurement.

In addition to generating data from the measure-ment of yield surfaces, we are also seeking toaddress the fundamental question of the proper defi-nition of plastic strain in the context of large inelas-tic deformations.3,4 We have successfully measuredyield surfaces that have moved so that they nolonger enclose the origin in stress space (for exam-ple, paths O-B2 and O-C1-C2 in Fig. 1).

In this situation, the material cannot beunloaded to zero stress without causing new plasticdeformations. The traditional way of defining plasticstrain is to identify it with the residual strainremaining when the load is removed. This definitionarose intuitively from consideration of uniaxial testswith small deformation, but it is clearly inadequatein the situation of more general states of loadingwhere the yield surface no longer encloses theorigin in stress space.

Plastic strain is not among the set of kinematicvariables that come from classical continuummechanics. Since it is a primitive variable in theconstitutive theory, one must be able to unambigu-ously identify it for the theory to be meaningfullypredictive and not simply a sophisticated curve-fit.

In view of Eq. 2, while the magnitude of theplastic strain increment depends on the strainincrement, its direction does not. The direction isthe same as the direction of the tensor, ρkl, whichdepends only on the current state and not on ratesof stress or strain. This fact will be used to verifythe validity of the prescription for identifying plastic strain.3,4

Two or more specimens will be loaded to anarbitrary inelastic state where the yield surface instress space does not enclose the origin. Thespecimens will then be given small plastic loadincrements starting from the same point on theyield surface, but having different directions instrain space. This is shown schematically in Fig. 3,in both (2-D) stress and strain space. The pointclosest to the origin on a given yield surface instress space is designated as Sp, and the corre-sponding state in strain space is identified as theplastic strain tensor, Ep.

The darkest yield surface in Fig. 3 represents aknown arbitrary state. The two lighter surfacesrepresent two subsequent yield surfaces obtainedfrom the first by loading in two different direc-tions. If the prescription3,4 is valid, the resultingplastic strain increments for the two cases shouldhave the same direction, although they may vary in magnitude. Thus, colinearity of the points

, , and in Fig. 3b would verify the prescription.

Modeling

We have begun efforts to model the anisotropicmaterial behavior exhibited in the experiments, and to explore issues related to numerical imple-mentation of anisotropic models. The Mises yieldcondition, which is a quadratic polynomial in the

E2pE1

pE0p


s22

(a)

(b)

s11

dS2

dS1

Sp0 Sp

1

Sp2

e22

e11

dE1

dE2

Ep0

Ep2

Ep1

Figure 3. Schematic of the experimental method for verifyingthe prescription for plastic strain viewed in (a) stress space,and (b) strain space. From a fixed arbitrary state (darkestyield surface), the change due to plastic increments on pathsof two differing directions is measured. Colinearity of thepoints , , and in strain space verifies theproposed prescription.

E p2E p

1E p0

605 Nikkel_qk 7/27/99 9:26 AM Page 6-4


deviatoric stress components, is known to agreewell with experimental data on annealed(isotropic) polycrystalline metals, and it also has aphysically appealing interpretation in terms ofdistortion energy.

For anisotropic materials, the most generalquadratic yield function representing a smoothinitial yield surface that reduces to the Mises yieldfunction in the special case of isotropic materials isof the form5

, (3)

where the coefficient tensor, Bklmn, has the obvioussymmetries, and hence has 21 independent compo-nents. If, as is commonly done, the further assump-tion is made that the yield behavior is independentof the mean stress (pressure), then the stress inEq. 3 can be replaced by the deviatoric stress, andthe coefficient tensor can be replaced by a reducedtensor which has 15 independent coefficients.

The yield function (Eq. 3) contains as a specialcase the anisotropic yield condition of Hill which isavailable in the DYNA code and is sometimes usedin sheet metal forming analyses. Even though Eq. 3is much more general than the Mises yield function,it still does not adequately represent the kinds ofmaterial behavior exhibited in Fig. 1. It does,however, provide a reasonable basis to begin explor-ing some of the issues associated with implementinganisotropic plasticity models into numerical codes.

To evaluate some of these issues, the yield func-tion (Eq. 3) was implemented into the parallelversion of the ALE3D code. Figure 4 shows theresults of the simulation of a thick-walledanisotropic sphere (initial radius = 10, initial wallthickness = 2) subjected to a uniform externalpressure load. With isotropic material properties,the sphere symmetrically compresses as onewould expect. By modifying the properties in onedirection, the very non-symmetric response inFig. 4 is predicted, indicating the significanteffect that material anisotropy can have on over-all structural response.

For Mises-type yield functions, an efficientnumerical procedure for integrating the evolutionequations has long been used. This consists of anelastic trial step followed by a radial-return correc-tor step. This simple and efficient procedure doesnot work in general for anisotropic yield functions,and one of the challenges ahead will be in develop-ing robust, numerically efficient procedures forintegrating the anisotropic evolution equations. Inthe context of the nonlinear strain-space formula-tion of the theory of elastic-plastic materials,

f B s sklmn kl mn – = κ 2

Papadopoulos and Lu6 have developed a method forintegrating the evolution equations, which for thespecial case of transversely isotropic materialsreduces to computing three separate radial-returnsteps for three orthogonal parts of the solution.

In addition to this effort to develop a phenomeno-logical continuum model using experimental data,an effort is also underway to develop a homogeniza-tion methodology to predict effective macroscopicbehavior based on explicit consideration ofmicrostructural features, such as the statisticaldistribution of grain orientations. To this end, amaterial model for single crystal plasticity has beenimplemented into NIKE3D. This model also has thecapability to account for polycrystal aggregates ateach integration point using Taylor averaging. Thismodel can be used in numerical experiments topredict effective yield surfaces for explicitmicrostructural configurations.

Future Work

We now have confidence in our ability to measureyield surfaces as accurately as necessary for ourpurposes. Continued work will involve generatingspecific data sets to guide the development of betterconstitutive equations for nonlinear plasticity. Theanisotropic yield function (Eq. 3), while a usefulstarting point for evaluating some of the basicnumerical issues, is not sufficiently general to repre-sent the behavior we are seeing in the data. We willbe focusing on developing a better theoretical modelto represent the yield surface and its evolution.

FY 98 6-5

-5

-5

5

0

5

5

5

-5

5

5

5

0

0

0

10

0

5

10

0

Figure 4. Numerical simulation of a sphere of homogeneousbut anisotropic material, subjected to a uniform externalpressure. For isotropic material properties, the sphere wouldcompact symmetrically.

605 Nikkel_qk 7/27/99 9:26 AM Page 6-5


We also plan to examine materials of particularrelevance to LLNL programs, relevance such astantalum, which is a target material for theMultiscale Material Modeling effort. We also plan tocomplete the experiments discussed above toaddress the question of the correct identification ofthe plastic strain tensor in the context of generalfinite deformations.

Acknowledgment

Contributions to this project from Prof. J. Caseyresult from work supported by a grant from theSolid Mechanics Program of the National ScienceFoundation. Part of support for A. Brown’s work,and material used during the development of theexperimental procedure, were also providedthrough this grant.

References

1. Naghdi, P. M. (1990), “A Critical Review of the Stateof Finite Plasticity,” J. Appl. Math. and Phys. (ZAMP),41, pp. 315-394.

2. Nikkel, D. J., A. A. Brown, and J. Casey (1998),“Evolution of Anisotropic Yield Behavior,”Engineering Research, Development and Technology,Lawrence Livermore National Laboratory, Livermore,California (UCRL-53868-97), pp. 5-1–5-5.

3. Casey, J., and P. M. Naghdi (1992), “A Prescriptionfor the Identification of Finite Plastic Strain,” Int. J.Engng. Sci., 30, pp. 1257-78.

4. Casey, J., and P. M. Naghdi (1992), “On theIdentification of Plastic Strain at FiniteDeformation,” in Defects and Anelasticity in theCharacterization of Crystalline Solids, L. M. Brock,ed., ASME AMD Vol. 148, pp. 11-33.

5. Green, A. E., and P. M. Naghdi (1965), “A GeneralTheory of an Elastic-Plastic Continuum,” Arch. Rat.Mech. and Anal., 18, pp. 251-281.

6. Papadopoulos, P. and J. Lu (1998), “A GeneralFramework for the Numerical Solution of Problems inFinite Elasto-Plasticity,” Comp. Meth. in Appl. Mech.and Engng., 159, pp. 1-18.


605 Nikkel_qk 7/27/99 9:26 AM Page 6-6

odeling Large-Strain High-Rate Deformation in Metals


Introduction

Many modeling problems of interest to LawrenceLivermore National Laboratory (LLNL) involveaccurate representation of the high-rate deforma-tion response of materials. Examples include themodeling of material processing operations as wellas the in-service performance of materials. Typicalmaterial processing operations, in which high-ratedeformation is observed, include material cutting,numerous forming operations (such as rolling andforging) and material polishing. Typical in-serviceperformance problems include the ballistic penetra-tion and perforation of armor materials, the perfor-mance of munitions, and explosive fragmentation.

Many of these problems are difficult to modelaccurately. Much of this difficulty arises from thelarge strains and adiabatic heat produced, which, inturn, causes increases in temperature with resultingchanges in material microstructure, material prop-erties, and deformation mechanisms. Large changesin strain rate are also produced.

In addition, deformation can produce instabilities inthe form of adiabatic shear bands. Voids can also beproduced that can influence flow behavior and serveas a precursor to fracture. Thus, accurate materialmodels are necessary for understanding deformationbehavior (and strength) as well as failure response.

Objectives

Material models that can adequately representthe deformation response during high-rate loadingmust account for large strains (and the resultingstrain hardening or softening), as well as largechanges in strain rate and temperature. Severalmodels have been developed that can represent, tovarying degrees, the high-rate deformationresponse of materials. Examples include models byJohnson and Cook (JC),1-3 Zerilli and Armstrong(ZA),4–6 and Follansbee and Kocks (mechanicalthreshold stress model)7.

Two of these models (JC and ZA) have been intro-duced into LLNL’s DYNA codes.

Of these two models, the JC model is much morewidely used. The JC model was developed during the1980s to study impact, ballistic penetration, andexplosive detonation problems. The model hasproven to be very popular and has been used exten-sively by a number of national laboratories, militarylaboratories, and private industry to study high-rate,large-strain problems. The reasons for the popular-ity of this model include the simple form of theconstitutive equations and the availability ofconstants used in the equations for a number ofmaterials. The JC material model also has a cumula-tive damage law that can be used to assess failure.

FY 98 6-7

The large-strain deformation responses of aluminum alloy 6061-T6 and titanium alloy Ti-6Al-4Vhave been evaluated over a range in strain rates from 10-4 s-1 to >104 s-1. The results have been usedto critically evaluate the strength and damage components of the Johnson-Cook (JC) material model.Two new models that address the short-comings of the JC model were then developed and evaluated.One model is derived from the rate equations that represent deformation mechanisms active duringmoderate- and high-rate loading; the other model accounts for the influence of void formation onyield and flow behavior of a ductile metal (the Gurson model). The characteristics and predictivecapabilities of these models are reviewed.

Donald R. Lesuer and Mary M. LeBlancManufacturing and Materials Engineering DivisionMechanical Engineering

Gregory J. KayNew Technologies Engineering DivisionMechanical Engineering

TA 610 Lesuer_qk 7/27/99 9:28 AM Page 6-7


In this report, we take a critical look at the JCmodel and its ability to represent the large-straindeformation behavior of two important structuralmaterials: an α-β titanium alloy (Ti-6Al-4V) and amoderate strength aluminum alloy (6061-T6). Themodel has been evaluated over a range of strainrates from 10-4 s-1 to >104 s-1. The damage law wasalso evaluated for its ability to predict failure inthese materials. Two new models were then devel-oped and evaluated that address some of the short-comings observed with the JC model. One of themodels is derived from the rate equations thatrepresent deformation mechanisms active duringmoderate- and high-rate loading; the other modelaccounts for the influence of void formation on yieldand flow behavior of a ductile metal (the Gursonmodel8). The characteristics and predictive capabili-ties of these models are reviewed.

Progress

Materials, Experiments and Results

The materials used in this study wereobtained from commercial sources. The 6061alloy was received as a hot, cross-rolled plate inthe T6 temper. The Ti-6Al-4V alloy was obtainedaccording to the AMS 4911 specification, whichproduced an equiaxed α and transformed β microstructure.

High-rate testing was done in both compressionand tension using the split Hopkinson pressure bartechnique, and data was obtained at strain rates of103 s-1 to 104 s-1. In the compression tests, thestrain histories for the incident and transmittedwaves in the elastic pressure bars were measuredand analyzed to determine the nominalstress/strain/strain-rate response of the sample. Inthe tension tests, the strain history in the elasticpressure bars was used to obtain the stress-timeresponse of the sample. The strain and strain-ratebehavior of the sample was obtained from high-speedphotographic images derived from a framing camera.

All stress-strain data is provided as “true stress”and “true strain.” The stress-strain data for 6061-T6aluminum obtained in tension and compression isshown in Fig. 1. The experiments in tension wereconducted at a strain rate of 8000 s-1, and sampleswere tested with the tensile axis parallel to thelongitudinal and transverse orientations in the plate.The experiments in compression were conducted ata strain rate of 4000 s-1, and samples were testedwith the compression axis parallel to the longitudi-nal, transverse, and through-thickness orientationsin the plate.

Data for “elastic” loading of the sample can notbe obtained in these tests due to wave propagationeffects. The stress-strain data shown in Fig. 1 canbe considered valid once the samples have yieldedplastically, which is accompanied by stress andstrain rate uniformity in the sample. The data for thedifferent orientations of testing show that in bothtension and compression, the stress-strain responseis highly isotropic. This is especially true incompression, in which the curves for the threeorientations fall virtually on top of one another. Thecompression samples deformed to the limits of theexperiment without failure, while the tensionsamples failed after a strain of 0.26

Johnson-Cook Material Model

The formulation for the JC model is empiricallybased and represents the flow stress with an equa-tion of the form

(1)

where σ is the effective stress, ε is the effective plas-tic strain, ε⋅* is the normalized effective plastic strainrate (typically normalized to a strain rate of 1.0 s-1),

is the homologous temperature, n is the workhardening exponent and A, B, C, and m are constants.

The values of A, B, C, n, and m are determinedfrom an empirical fit of flow stress data (as a function

T*

σ ε ε= +

+

A B C Tn m1 1ln ˙ – ,* *


0 0.1 0.2 0.3 0.4 0.5 0.6Strain

0

100

200

300

400

500

Stre

ss (

MPa

)

Necklongitudinalspecimen

Necktransversespecimen

Tension - transverse direction

6061-T6

Tension - longitudinal direction

Compression - longitudinal, transverseand through thickness directions

Figure 1. Stress-strain data for aluminum alloy 6061-T6obtained in tension and compression with the split Hopkinsonpressure bar apparatus. The experiments in tension wereconducted in two different orientations and at a strain rate of8000 s-1. The experiments in compression were conducted inthree different orientations and at a strain rate of 4000 s-1.



of strain, strain rate and temperature) to Eq. 1. Forhigh-rate deformation problems, we can assume thatan arbitrary percentage of the plastic work doneduring deformation produces heat in the deformingmaterial. For many materials, 100% of the plasticwork becomes heat in the material. Thus the temper-ature used in Eq. 1 can be derived from the increasein temperature according to the following expression

(2)

where ∆T is the temperature increase, α is thepercentage of plastic work transformed to heat, c isthe heat capacity and ρ is the density.

Fracture in the JC material model is derived fromthe following cumulative damage law:

(3)

where

(4)∆ε is the increment of effective plastic strain duringan increment in loading and is the mean stressnormalized by the effective stress. The parametersD1, D2, D3, D4, and D5 are constants. Failure isassumed to occur when D = 1. The current failurestrain (εf) is thus a function of mean stress, strainrate, and temperature. The constants for the JCmodel used in the evaluations in the next section aregiven in Table 1.

Model Evaluation. The adiabatic stress-strainbehavior for the 6061-T6 alloy predicted by the JCmaterial model is shown in Fig. 2 for loading intension, compression, and shear. The cumulativedamage predicted by the failure model is alsoshown in the figure, and the failure strains for thethree stress states are indicated on the stress-strain curve. The three stress states show differentdamage curves because of the influence of themean stress term on εf in Eq. 4.

The stress-strain response predicted by the mater-ial model is compared against the experimental datain both tension and compression in Fig. 3. The yield

σ*

ε σ εf D D D D D T= +

+

+

1 2 3 4 51 1exp ln ˙ ,* * *

D

f= ∑ ∆ε

ε,

∆TB

c nn=

+( )+α

ρε

11,

strength predicted by the JC model correlates verywell with the experimental results. However, theexperimental stress-strain curves work harden at ahigher rate. This is not a fundamental short-comingof the model, since higher work hardening rates arepossible with larger values of B and n in Eq. 1. Thefailure strain in tension as predicted by the JC mater-ial model (εf = 0.52) is significantly higher than thatobtained experimentally (εf = 0.26). This is a signifi-cant difference and the physical origins of thisdiscrepancy need to be understood. However,detailed studies of failure models are outside thescope of this paper.

The stress-strain rate response for the 6061-T6alloy is compared against the predictions of the JCmodel in Fig. 4. Data was obtained from the work ofNicholas9 as well as from this study. Here significantdeviations between model predictions and experimen-tal results are evident. The experimental data shows adramatic increase in strength above a strain rate of103 s-1. This increase in strength has been observed ina number of metals10 and is generally recognized asresulting from a change in deformation mechanism.

FY 98 6-9

Table 1. Johnson-Cook constants for Ti-6Al-4V and 6061-T6.

A B n C m D1 D2 D3 D4 D5(MPa) (MPa)

6061-T6 324 114 0.42 0.002 1.34 -0.77 1.45 -0.47 0.0 1.60Ti-6Al-4V 862 331 0.34 0.012 0.8 -0.09 0.25 -0.5 0.014 3.87

0.2

0.4

0.6

0.8

Dam

age

400

50

100

150

200

250

300

350

1.20

0 0.2 0.4 0.6 0.8 1.0Strain

STRESS

DAMAGE

Tension

Shear

Compression

T SC

Stre

ss (

MPa

)

1.0

0

Figure 2. Adiabatic stress-strain behavior for aluminum alloy6061-T6 at a strain rate of 6000 s,-1 predicted by the JC materialmodel. Results are presented for loading in tension, compression,and shear. The cumulative damage predicted by the materialmodel is also shown. The failure point along the stress-straincurve is shown for tension, shear, and compression.



At lower strain rates, the deformation rate iscontrolled by the cutting or by-passing of discreteobstacles by dislocations. At higher rates, the defor-mation rate is controlled by phonon or electron dragon moving dislocations. These two mechanisms arerepresented by different deformation rate equations,which results in the dramatic change in behaviorfrom low strain rates to high strain rates. Suchdramatic changes are outside the scope of the JCmodel. In the next section we present a model thataccounts for these mechanism changes.

The predictions of the JC model for the Ti-6Al-4Valloy are shown in Fig. 5 and compared againstexperimental data in Figs. 6 and 7, obtained fromthe work of Wulf11, Meyer12 and Follansbee andGray13. The same capabilities and limitations of thematerial model that were observed for the 6061-T6

alloy were noted for the Ti-6Al-4V alloy. The modelcan adequately represent work-hardening behaviorin both materials. The most serious limitation is itsability to predict variations of flow stress withstrain rate, as shown in Fig. 7. The failure modelpredicted the correct ductility in tension for theTi-6Al-4V alloy (εf = 0.15) but, in compression,the model predicted a significantly higher ductilitythan that observed experimentally.

Mechanism-Based Material Model

Rate Equations. We now derive a rate equationrepresenting deformation that can be controlled bytwo sequential processes: 1) the cutting (or by-passing) of obstacles by dislocations, or 2) the dragon moving dislocations by phonons or electrons. The


0 0.2 0.4 0.6 0.8 1.0 1.2Strain

0

100

200

300

400

500

Stre

ss (

MPa

)

Tension (8000 s–1)Compression (4000 s–1)Johnson-Cook model (6000 s–1)adiabatic stress-strain response

Shearεf = .8

Compressionεf = 1.15

Tensionεf = .52

Figure 3. Comparison between the stress-strain behaviorpredicted by the JC material model and experimental data foraluminum alloy 6061-T6.

10–4 10–3 10–2 10–1 100 101 102 103 104

Strain (s–1)

300

350

400

450

500

Stre

ss (

MPa

)

ε = 0.2

ε = 0.1

Figure 4. Comparison between the stress-strain rate behaviorpredicted by the JC material model and experimental data foraluminum alloy 6061-T6. Experimental data is derived from acompilation of LLNL results and data from Reference 9.

340

360

380

400

Tem

per

atur

e (K

)

1200

200

400

600

800

1000

0.300

0 0.05 0.10 0.15 0.20 0.25Strain

Stre

ss (

MPa

)

420

280

300

320Temperature

Stress

Tensionε

f = .0146

Shear εf = 0.196

Compressionε

f = 0.258

Figure 5. Adiabatic stress-strain behavior for titanium alloyTi-6Al-4V at a strain rate of 5000 s-1, as predicted by the JCmaterial model. Results are presented for loading in tension,compression, and shear. The adiabatic temperature rise is alsoshown. The failure point along the stress-strain curve is shownfor the three stress states.

0 0.05 0.10 0.15 0.20 0.25 0.30Strain

0

200

400

600

1000

1400

800

1200

1600

Stre

ss (

MPa

)

Johnson-Cook model (5000 s–1)adiabatic stress-strain responseWulff - compression (8000 s–1)Meyer - compression (2000 s–1)Follensbee - tension (3000 s–1)

Figure 6. Comparison between the stress-strain behaviorpredicted by the JC material model and experimental data fortitanium alloy Ti-6Al-4V (obtained from References 11 to 13 ).



problem is illustrated schematically in Fig. 8, whichshows dislocations in contact with discrete obsta-cles that have an average spacing, d. After Frost andAshby14, the average velocity, v, for a dislocationmoving through these obstacles is

(5)

where t1 is the time required to cut or by-pass theobstacle, and t2 is the time spent moving to the nextobstacle. Different rate equations represent thedeformation kinetics associated with discrete obsta-cles and drag. Let ε⋅1 represent the strain rate whendeformation is controlled by the cutting or by-passing of discrete obstacles, and ε⋅2 represent thestrain rate when deformation is controlled by dragon moving dislocations. Since

(6) ˙ ,ε ρ= bv

v d t t= +( )/ ,1 2

where ε⋅ is the strain rate, b is the Burger’s vectorand ρ is the mobile dislocation density,

(7)

where ε⋅eff is the effective strain rate on the slipplane shown in Fig. 8. Thus

(8)

The rate equation for discrete obstacle controlledplasticity15 can be taken as

(9)

where ε⋅0 is a constant, Q is an activation energy, k isBoltzmann’s constant, σ is the stress and τ is thestrength of the obstacle. At constant temperature,the equation can be taken as

(10)

where A and B are constants. The rate equation forphonon- or electron-drag-controlled plasticity can betaken as

(11)

where C and D are constants. Several theoreticaltreatments have shown that D approaches 116,17.We will use the general form of the rate equationshown in Eq. 11. Equations 8, 10, and 11 can nowbe used to calculate the strain rate resulting fromthe sequential mechanism of discrete-obstacle plas-ticity and drag-controlled plasticity.

Model Evaluation. The model, as representedby Eqs. 8, 10, and 11, was evaluated against thestress-strain rate data for the 6061-T6 and Ti-6Al-4V alloys shown in Figs. 4 and 7, respectively. Theconstants for obstacle-controlled plasticity (A andB) were evaluated in the strain-rate range wherethis mechanism is dominant. Similarly, theconstants for drag-controlled plasticity (C and D)were evaluated in the strain-rate range where thismechanism is dominant.

Figure 9 shows a comparison of the stress-strain rate response predicted by the mechanism-based material model and experimental data forthe 6061-T6 alloy.

Similarly, Fig. 10 shows a comparison of thestress-strain rate response predicted by the mech-anism-based material model and experimental

˙ ,ε σ2 = C D

˙ exp ,ε σ1 = ( )A B

˙ ˙ exp – – ,ε ε στ1 0 1=

QkT

˙

˙ ˙˙ ˙

.ε ε εε εeff =

+1 2

1 2

˙

˙ ˙

,ε

ε ε

effd

d d=

+1 2

FY 98 6-11

10–5 10–3 10–1 101 103 105

Strain (s–1)

1500

1400

1300

1200

1100

1000

900

800

700

600

500

Stre

ss (

MPa

)

+ LLNL data (1400 MPa @ 5000 s–1)o Handbook data (1000 MPa @ 10-4 s–1)

o

Johnson-Cook model (ε=0.04)Meyer-tension (ε=0.002)Meyer-compensation (ε=0.002)Follensbee/Gray (ε=0.04)Wulf (ε=0.1)

+

Figure 7. Comparison between the stress-strain rate behaviorpredicted by the JC material model and experimental data foraluminum alloy 6061-T6.

d

Dislocationposition 1

Dislocationposition 2

Slip plan

Figure 8. Dislocations on a slip plane in contact with discreteobstacles. The shear stress on the slip plane is σ, and the aver-age spacing between obstacles is d. At high strain rates, thedislocation velocity (and therefore strain rate) is determined bythe rate at which the discrete obstacles are by-passed, or therate at which the dislocation moves from one discrete obstacleto the next.



data for the Ti-6Al-4V alloy. The figures also showthe regions of the stress-strain curves that aredominated by discrete-obstacle plasticity and bydrag-controlled plasticity. For both alloys, the agree-ment between the model predictions and experimen-tal data is excellent.

Gurson Void Growth Model

Observations have been made that ductile frac-ture in metals is related to the nucleation andgrowth of voids. Conventional plasticity models, forexample, von Mises, are based on the assumptionof plastic incompressibility and can not predict thegrowth of voids during yielding. Studies have indi-cated18–20 that void growth during tensile loadingis related to the hydrostatic component of stress,and that this porosity increase directly affectsmaterial yielding.

In these observations it was assumed that thematerial surrounding a void was incompressible.Gurson8 proposed a pressure-sensitive macroscopicyield surface that relates void growth to the evolu-tion of microscopic (pointwise physical quantities ofthe matrix material) and macroscopic quantities toaccount for the behavior of void-containing solids.Here, macroscopic refers to the average values ofphysical quantities, which represent the materialaggregate behavior. As defined by Gurson, the yieldsurface for a ductile material is:

(12)where σo is the tensile flow stress of the micro-scopic matrix material, q and p are the equivalent

Φ =

+

+

=q

q fp

q fσ σ0

2

10

222

32

1 0cosh – ,

stress and hydrostatic stresses of the macroscopicmaterial, and f is the current void volume fractionwhich is a function of the initial porosity, the voidgrowth, and nucleation during yielding. The materialparameters q1, q2 are defined by Gurson.

The Gurson model was added to NIKE2D byB. Engelmann. For the current study, a version ofthe NIKE2D Gurson model was modified to correctlyaccount for the evolution of plastic strain in themicro (matrix) material and to account for strainrate sensitivity. The model was added to DYNA3D.

The response of a notched bar under uniaxialtensile loading was simulated to demonstrate theDYNA3D application of the Gurson model.Substantial hydrostatic tension is created in thenotched regions of the bar for this type of loading.This hydrostatic stress accelerates void growth andleads to the eventual coalescence of voids andductile failure of the bar. Failure was assumed tocorrespond to the loss of load-carrying capability inthis displacement-controlled simulation.

The bar was assumed to have the following mater-ial properties: E = 20.7 GPa, υ = 0.3, yield stress =690 MPa, with a linear hardening modulus of 1,540MPa. The initial void fraction was assumed to beequal to 0.050.

The initial and deformed shapes of the tensilespecimen are shown in Fig. 11, which also depictsthe regions of predicted high void growth. The effectof rate-dependence is shown in Fig. 12, where anincreased loading rate resulted in an increasednormalized axial load (actual axial load/initialyield strength), with softening similar to the rate-independent Gurson model results.

Also shown in Fig. 12 is the conventional plas-ticity solution, which does not exhibit the


10–4 10–2 100 102 104

Strain rate (s–1)

300

350

400

450

500

Stre

ss (

MPa

)

Drag-control

PredictionData

Discrete-obstacle-controlled

Figure 9. Comparison between the stress-strain rate behav-ior predicted by the mechanism-based material model andexperimental data for aluminum alloy 6061-T6. Regions ofthe stress-strain rate curve that are dominated by discrete-obstacle plasticity and drag-controlled plasticity are shown.

10–310–4 10–110–2 101100 103 104102

Strain rate (s–1)

1500

1400

1300

1200

1100

1000

900

Stre

ss (

MPa

)

Drag-controlled

PredictionData

Discrete-obstacle-controlled

Figure 10.Comparison between the stress-strain rate behaviorpredicted by the mechanism-based material model andexperimental data for titanium alloy Ti-6Al-4V. Regions ofthe stress-strain rate curve that are dominated by discrete-obstacle plasticity and drag-controlled plasticity are shown.



pronounced softening predicted by the Gursonmodel. The conventional plasticity yield surface isalso shown to be larger, with a higher strain-to-failure, than the porous material, a resultconfirmed by experimental results.

For this simulation, the final void fraction was0.70. A calculation was also performed to checkthe sensitivity of the solution to mesh size. Themesh in this calculation was twice the density ofthe initial simulation. The results of this calcula-

FY 98 6-13

Uniform axial displacementapplied to upper and lowerboundaries.

Original mesh

Deformed mesh

Initial voidfraction equalto 0.050

Final maximumvoid fractionequal to 0.70

(a) (b) Figure 11.Notchedtensile specimen voidgrowth, as predictedby the Gurson model.

0 0.05 0.10 0.15 0.20Average axial strain

0

0.049

0.097

0.194

0.291

0.146

0.243

0.340

Axi

al lo

ad/I

nit

ial y

ield

str

eng

th Strain rate

equal to 2,921 s–1Rate-dependentGurson model

Rate-independentGurson model

Conventional plasticityresults

Figure 12.Gurson model rate effects for a notched bar underdisplacement-controlled axial loading.

0 0.05 0.10 0.15 0.20Average axial strain

0

0.030

0.060

0.090

0.120

0.180

0.240

0.150

0.210

0.270

Axi

al lo

ad/I

nit

ial y

ield

str

eng

th

Coarse mesh results (mesh as shown in Figure 1).

Fine meshresults(2X numberof elements)

Conventional plasticity results

Figure 13.Gurson model mesh sensitivity for a notched barunder displacement-controlled axial loading.



tion, shown in Fig. 13, indicate that there issome small mesh sensitivity of the solution, inthe post-failure phase, for the rate-independentGurson solution.

Summary

The primary conclusions and observations rela-tive to the three models studied are as follows:

1. JC model. For the alloys studied, the JC modelcan accurately represent the yield and work-hardening behavior of the materials. The JCmodel predicts higher failure strains than thoseobserved experimentally. The most seriousshort-coming of the JC model is its inability toaccurately represent the variation of flow stresswith strain rate.

2. Deformation mechanism model. This modelaccounts for two sequential deformation mecha-nisms that are active at moderate- and high-deformation rates. The mechanisms arediscrete-obstacle plasticity and drag-controlledplasticity. The model has been developed andevaluated against stress-strain rate data for the6061-T6 alloy and theTi-6Al-4V alloy. Agreementbetween experimental results and model predic-tions is excellent.

3. Gurson void growth model. The Gurson voidgrowth model has been introduced into theDYNA3D code. The model was modified toaccount for the evolution of plastic strain andstrain rate sensitivity. The model was used inthe DYNA3D code to simulate the response of anotched bar during tensile loading.

References

1. Johnson, G. R., and W. H. Cook (1985), “FractureCharacteristics of Three Metals Subjected toVarious Strains, Strain Rates, Temperatures andPressures,” Engineering Fracture Mechanics, 21,No. 1, pp. 31–48.

2. Johnson, G. R., and W. H. Cook (1983), “AConstitutive Model and Data for Metals Subjectedto Large Strains, High Rates and HighTemperatures,” Proceedings of the SeventhInternational Symposium on Ballistics, The Hague,The Netherlands, pp. 541–547.

3. Johnson, G. R., and T. J. Holmquist (1988),“Evaluation of Cylinder-Impact Test Data forConstitutive Model Constants,” Journal of AppliedPhysics, 64, No. 8, pp. 3901–3910.

4. Zerilli, F. J., and R. W. Armstrong (1987),“Dislocation-Mechanics-Based Constitutive Relationsfor Material Dynamics Calculations,” Journal ofApplied Physics, 61, No. 5, pp. 1816–1825.

5. Zerilli, F. J., and R. W. Armstrong (1990),“Description of Tantalum Deformation Behavior byDislocation Mechanics Based ConstitutiveEquations,” Journal of Applied Physics, 68, No. 4,pp. 1580–1591.

6. Zerilli, F. J., and R. W. Armstrong (1992), “The Effectof Dislocation Drag on the Stress-Strain Behavior ofFCC Metals,” Acta Metallurgica et Materialia, 40, No.8, pp. 1803–1808.

7. Follansbee, P. S., and U. F. Kocks (1988), “AConstitutive Description of the Deformation ofCopper Based on the Use of the MechanicalThreshold Stress as an Internal State Variable,” ActaMetallurgica, 36, pp. 81–93.

8. Gurson, A. J. (1977), “Continuum Theory ofDuctile Rupture by Void Nucleation and Growth,Part I - Yield Criteria and Flow Rules for PorousDuctile Media,” Journal of Engineering MaterialsTechnology, 99, pp. 2–15.

9. Nicholas, T. (1981), “Tensile Testing of Material atHigh Rates of Strain,” Experimental Mechanics, May,pp. 177–185.

10. Follansbee, P. S. (1986), “High-Strain-RateDeformation of FCC Metals and Alloys,” inMetallurgical Applications of Shock-Wave and High-Strain-Rate Phenomena, L. E. Murr, K. P.Staudhammer, and M. A. Meyers, eds. New York:Marcel Dekker, Inc., pp. 451–479.

11. Wulf, G. L. (1979), “High Strain Rate Compression ofTitanium and Some Titanium Alloys,” InternationalJournal of Mechanical Sciences, 21, pp. 713–718.

12. Meyer, L. W. (1984), “Strength and Ductility of aTitanium-Alloy Ti-6Al-4V in Tensile and CompressiveLoading Under Low, Medium and High Rates ofStrain,” in Titanium Science and Technology, G.Lutjering, U. Zwicker, and W. Bunk, eds., DeutscheGessellschaft fur Metallkunde .

13. Follansbee, P. S., and I. G.T. Gray (1989), “AnAnalysis of the Low Temperature, Low and HighStrain-Rate Deformation of Ti-6Al-4V,” MetallurgicalTransactions A, 20A, pp. 863–874.

14. Frost, H. J., and M. F. Ashby (1971), “Motion of aDislocation Acted on by a Viscous Drag Through anArray of Discrete Obstacles,” Journal of AppliedPhysics, 42, No. 13, pp. 5273–5279.

15. Frost, H. J., and M. F. Ashby (1982), DeformationMechanism Maps, Pergamon Press, Oxford.




16. Granato, A. V. (1973), “Microscopic Mechanisms ofDislocation Drag,” Metallurgical Effects at HighStrain Rates, Plenum Press, New York, New York,pp. 255–275.

17. Kumar, A., F. E. Hauser, and J. E. Dorn (1968),“Viscous Drag on Dislocations in Aluminum at HighStrain Rates,” Acta Metallurgica, 16, pp. 1189–1197.

18. McClintock, F. A. (1968), “A Criterion for DuctileFracture by the Growth of Holes,” Journal of AppliedMechanics, 35, pp. 363–371.

19. Rice, J. R., and D. M. Tracy (1969), “On the DuctileEnlargement of Voids in Triaxial Stress Fields,”Journal of the Mechanics and Physics of Solids, 17,pp. 201–217.

20. Kahlow, K. J., and B. Avitzur (1969), “VoidBehavior as Influenced by Deformation andPressure,” American Iron and Steel Institute,Lehigh University.

FY 98 6-15


niform Etching of 85-Cm-Diameter Grating


Introduction

During a joint effort between laser and engineer-ing personnel at Lawrence Livermore NationalLaboratory (LLNL), an ion beam etching systemhaving a 40-cm diameter gridded, broad-beam ionsource was established. At the beginning of thisproject, this system was capable of etching a 30-cmpart with an etch depth uniformity of better than±5%. Our goal was to extend our etching technologybase to allow us to etch larger parts in the samesystem, with the ultimate goal of producing 85-cm-diameter transmission gratings with high efficiencyacross this full diameter. To accomplish this, weneeded to develop a shadow mask allowing us toexactly cancel the inherent non-uniformity in the ionbeam. We also needed to determine the etchingparameters that are most critical for generating thedesired grating profiles to minimize the risk of anunsuccessful etch on a large part.

Progress

Ion Source Modification

To make possible the uniform etching of suchlarge parts, we needed to modify the existing ion

source to broaden the ion beam beyond the rangefor which it was designed. We first removed the orig-inal equipment internal baffle to permit use of theentire ion beam. We next removed the accelerationgrids and re-installed them in a convex configurationto provide a de-focussed beam with a larger effectivediameter. The result of these two modifications canbe seen in Fig. 1. While the beam intensity is dimin-ished, its diameter is increased. We needed thisadditional beam diameter to make it possible to etchour large part uniformly.

Substrate Holder and Etching Geometry

In addition to the ion source modifications, thesubstrate stage was modified to accept a single 85-cm-diameter substrate in place of the existing four30-cm substrates. This was done while maintainingthe ability to control substrate temperature.

Etch geometry, pictured in Fig. 2, was dictatedby existing hardware: ion source and substratestage horizontally opposed, with centers offset by22 cm. The large disk which holds the 85-cmsubstrate is rotated with the ion source mountedoff-center so that the entire surface is exposed tothe ion beam during some portion its rotationabout its center axis.

FY 98 6-17

The purpose of this project was to extend the capability of an existing gridded, broad-beam(Kaufman-style) ion beam system to permit the uniform etching of a fused silica optic 85 cm in diam-eter with an etch depth uniformity within 5%. Since we hoped to demonstrate that we would be ableto fabricate large size diffractive optics (for example, transmission gratings), a secondary require-ment was to establish the etching conditions that allow such grating structures to have adequate effi-ciencies over this same large area.

We had hoped, during the course of this project, to fabricate a full size 85-cm diameter part. Sincethe substrate blanks for such gratings cost $50,000 each and most of that cost would be sacrificedduring an unsuccessful etch, we considered it important to demonstrate a reproducible andpredictable process prior to risking such a blank. At this time, we believe we have demonstrated thatwe can etch such a part with minimal risk of failure. This was accomplished by etching a series ofsmall gratings and fused silica etch witnesses mounted across this diameter, the data from which arepresented in this report.

Steven R. Bryan, Jr. and David L. SandersManufacturing Materials Engineering DivisionMechanical Engineering

615 Bryan_qk 7/27/99 9:32 AM Page 6-17


Model to Define Shadow Mask

To improve etch uniformity, a beam-shaping bafflewas placed between the ion source and thesubstrate, as near to the substrate as possible. Thedetermination of the shape of this baffle, which ulti-mately determines the etch uniformity, was criticalto meet the goals of this project.

Dependable representation of the actual etchuniformity by a model is dependent upon consis-tent, reproducible operation of the ion source.Concerns about ion source repeatability were alle-viated by periodically mapping the ion currentdensity along a radius of the ion beam. As seen inFig. 3, the current density profile of the ion sourcewas found to correlate quite well to fused silicaremoval (etch) rates made at the same distancesfrom the source. Ion source parameters, such asbeam voltage, beam current, accelerator voltage,discharge voltage, and flow rates for various reac-tive gases were defined previously and heldconstant throughout this project.

The etch model was established by:1. establishing the relationship between etch rate

and position relative to the ion source;

2. identifying a second equation that describes themotion of any point on the substrate (relative tothe ion source) as a function of time; and

3. substituting the position equation into the etchrate equation to yield etch rate as a function oftime for any given point on the substrate.

By performing a numerical integration of etchrate over the period of time that the point on thesubstrate is in the ion beam, one can compute theetch depth at that point.

Initially, etch rates were measured by position-ing fused silica samples in the ion beam, etchingthe sample for a period of time, then computingthe etch rate by dividing the measured etch depthon the sample by the etch time. After the relation-ship between etch rates and beam current densitymeasurements was established, a more economi-cal method became available: ion current densitymeasurements could be used to accurately predictetch rates. When confined to a plane that containsthe surface of the substrate (27.5 cm from the ionsource) and the perimeter of the ion beam, theradially symmetric beam yielded a fused silicaetch rate that was found to conform well to thepolynomial relation:


0.06

0.00

0.05

0.04

0.02

0.03

0.01

0.06

0.00

0.05

0.04

0.02

0.03

0.01

Convex grid

Concave grid

0.00 45.0040.0035.0030.0025.0020.0015.0010.005.00

0.00 45.0040.0035.0030.0025.0020.0015.0010.005.00

Ion

cur

ren

t d

ensi

ty (

mA

/cm

2 )Io

n c

urre

nt

den

sity

(m

A/c

m2 )

Distance from ion source center (cm)

Distance from ion source center (cm)

27.5 cm

27.5 cm

11 cm

Figure 1. Effect on ion source beam diameter resulting from the source modification described in the text.

615 Bryan_qk 7/27/99 9:32 AM Page 6-18


, (1)

where E(r) is etch rate (µm/h) as a function ofradial distance, r (cm), from the center of the ionsource. From this equation, one can immediately

+ ⋅

⋅

+ ⋅

.

– .

.

–

–

–

3 06 10

1 79 10

2 72 10

3 2

4 3

6 4

r

r

r

E r r( ) . – . –= ⋅1 42 4 41 10 4

see that the etch rate in the center of the ionsource, at a distance of 27.5 cm is 1.42 µm/h(Eq. 1 and Fig. 3).

The distance between a point on the substrateand the ion source center at any given time is depen-dent upon two variables: the location of the point onthe substrate and the angle of substrate rotation atthe time the distance is measured. Applying thePythagorean theorem to the etch geometry in Fig. 4yields the simple relation: r2 = x2 + y2 (where r isdistance from the center of the ion source).Substituting for variables x and y, the equivalent interms of substrate rotation angle, θ, and distancebetween substrate center and the point of interest,rs (Fig. 4), yields:

.

Since the rotational angle is a linear function oftime (t), one may substitute time for angle (θ),simplify the equation, and solve for only the positiveroot to produce a usable function:

. (2)

Note that points on the substrate have beendefined only by their distance from the center of thesubstrate. This is acceptable because all points on a

r r t r t rs s s , – cos( ) ( ) = +484 44 2

r r rs s

2 2 222 – cos sin = ( ) + ( )θ θ

FY 98 6-19

Substrate surface

Side view

View from ion source

Etch area(limits of

etch integration)

Point of substrate

Substraterotation

22 c

m

Ion source

Figure 2. Etch geometry. The geometry was defined by existinghardware with the ion source and substrate horizontallyopposed and the axes offset by 22 cm. The source-to-substratedistance was set to 27.5 cm.

16000

0

2000

8000

10000

12000

14000

6000

4000

0.03

0

0.025

0.02

0.015

0.01

0.005

Polynomial regressionEtch depth

Ion current density

Notes:1. 40-cm ion source fitted with convex grids2. All measurements made 27.5 cm from source

Ion

cur

ren

t d

ensi

ty (

mA

/cm

2 )

Center of 40-cm diameter ion source

454035302015105 250Radial distance from center (cm)

E(r) = 1.42 – 4.41 . 10-4 r + 3.06 . 10-3 r2 – 1.79 . 10-4 r3 + 2.72 . 10-6 r4

Etch

rat

e (Å

/h)

Figure 3. Ion beam profile comparison. Satisfactory correlation between Faraday cup measurements and fused silica etch rateswere observed.

615 Bryan_qk 7/27/99 9:32 AM Page 6-19


given circle of radius, rs, on the substrate will beexposed to the same ion flux profile from the ionsource, and will etch the same amount.

Substituting the equation defining radial distancefrom the source (Eq. 2) into the equation of etchrate (Eq. 1) above yields etch rate as a function oftime for any point on the substrate. Integration ofthis etch rate over a period of time will provide avery close approximation of the total etch depth:

, (3)

where rs is the location of any point on the substrateand θlimit is a new variable; the upper limit of integra-tion. This upper limit of integration, defined by theshape of a shadow-mask placed directly in front of thesubstrate (refer to Fig. 5), controls etch uniformityover the surface of the substrate. Increasing the valueof θlimit increases the amount of etch time on a revolu-tion of the substrate, thus increasing etch depth on aparticular circle of points on the substrate. Similarly,reducing the value of θlimit will reduce etch depth forthe same circle of points. As one can imagine, solvingthe resulting equation is tedious if not particularlycomplex, but is well suited to numerical integrationby computer.

Figure 6 shows the effect of the shadow mask onthe uniformity of the etching. In the case without themask, the etching rate varies by almost a factor of 7over the part. With a properly designed mask this

D r E

r t r dt

s

s s

,

– cos( )

θ θlimit

limit( ) =

+

∫02484 44 x

variation is seen to be within the goal of ±5% overmost of the part. (There remains a small area lessthan 2 cm at the center of the part that is slightly outof tolerance, but we are confident that this can becorrected with a minor adjustment of the position ofthe mask if such an adjustment is deemed necessary).

Grating Structures

As mentioned in the introduction, in addition to thenecessary uniformity in etching rate, it will be neces-sary to etch steep walled grating structures toachieve the required diffraction efficiency. Figure 7shows a scanning electron microscope photograph ofa fracture surface of such a grating structure. Similarprofiles were observed across the entire width of thegrating, indicating that the desired etching behaviorcan be obtained with our current etching parameters.

Other Important Process Parameters

Initially, based on the experience of others, weassumed that it would be necessary to cool thepart being etched to avoid overheating thephotoresist. (Such overheating can make thephotoresist difficult to remove after the comple-tion of the etching process). To accomplish thiscooling, we designed the substrate holder arounda thermally conductive dry-chuck materialnormally used for cooling substrates during ion-etching operations in the semiconductor industry.For our particular application, however, we


Point on substrate

Substrate rotation

Center of substrate rotation

Center of ion source

85-cm substrate

22.000 cm

x

ry

rs

Θ

Point on substrate

Open area

Ion sourcecenter

Masked area

Mask outline:defines upper limit ofintegration

Center of substraterotation

85-cm substrate22.000 cm

r rs

Θ

Figure 4. Location of point on substrate. Note that all pointscan be defined in terms of radial position on substrate. θ is alinear function of time.

Figure 5. Approximate position and shape of limit-definingshadow mask.

615 Bryan_qk 7/27/99 9:32 AM Page 6-20


discovered that cooling to temperatures belowroom temperature was not desirable because itcontributed to the build-up of a hydrocarbondeposit in the trenches of the grating. In fact, wefound it necessary to allow the substrate to beheated by the ion beam to prevent such buildup.In the future, we will need to investigate thistemperature effect in more detail to determine theoptimal etch temperature.

A second somewhat unexpected effect was theeffect of buildup of a slightly conductive film on theion gun insulators during the etching process. Thisbuildup was found to lead to a reduction of the ion(etching) current during the etching run by as muchas 20%. Recognizing this effect allowed us tocompensate by adjusting the ion source to maintaina constant ion etching current throughout the courseof the run.

Future Work

As funding and time permit, we plan to etch an85-cm fused silica grating optic using the parame-ters we have established during this project. Basedon our measurements, we expect the resulting grat-ing will have the required etch depth uniformity andgrating profile over the full diameter.

FY 98 6-21

16000

0

2000

8000

10000

12000

14000

6000

4000

454035302015105 250Radial distance from substrate center (cm)

Circumference of 85-cm diameter substrateCenter of 85-cm diameter substrate

Etch

dep

th (

Å)

Etch profile w/o beam shaping mask

Etch profile w/ beam shaping mask

Figure 6. Effect of the shadow mask pictured in Figure 5 on the etch uniformity across an 85-cm-diameter part. Since the part isrotated, this figure represents half of the full diameter.

Figure 7. SEM micrograph of a grating structure etched usingthe current system.

615 Bryan_qk 7/27/99 9:32 AM Page 6-21

istributed Sensor Inertial Measurement Unit


Introduction

Spacecraft motions are typically measured byIMUs capable of six degrees of freedom (that is,linear and angular motions about three orthogonalaxes). Typically, these units consist of integral pack-ages or sensors which are located at, or in the vicin-ity of, the CG. Our project addresses the need tokeep the vicinity of the CG free from IMU equipment,and to avoid the excessive weight of gyroscopes.Thus, we have developed an IMU system that usesonly accelerometers, none of which are at the CG.

The theory of accelerometer-only IMU systems isbased on the relation

(1)

where a is the accelerometer response; Acm is theacceleration of the CG; R is the accelerometer loca-tion relative to the CG; n is its sensing direction; ωis the angular velocity vector of the body; and isthe angular acceleration vector.

All the vectors are in the rotating frame of thebody. For conventional IMUs, three gyroscopes areused to give the three components of ω directly.Then three accelerometers, with mutually perpen-dicular sensing directions, give the informationneeded to find Acm, since the angular accelerationand centripetal acceleration terms in the equationfor a (the second and third terms) can be estimatedfrom the gyroscopic data and subtracted.

ω

a = ⋅ + ⋅ × + ⋅ × ×n A n R n Rcm ω ω ω

The rest of the navigation problem, as discussedby Regan and Anandakrishnan,1 is to integrate theangular rates over time and find the true orientationof the body in space at each instant. The body-frameacceleration can then be transformed to the inertial-frame acceleration, which is then itself integrated togive the true velocity and location of the body.

It is possible to determine the complete motion ofa body from acceleration measurements only. Usingnine or more accelerometers in different locationswith different sensing directions, Acm, ω, and canbe determined simultaneously. Various methodshave been proposed.2 The number of accelerometersis reduced to six if ω is found by integrating . Thisprocedure can be numerically unstable.2

Our design, based on the paper by Chen, Lee andDe Bra,3 chooses a special set of locations andorientations for six accelerometers, for which isobtained independently of the current value of ω,avoiding numerical instability. In this design, the sixaccelerometers are placed at the centers of the sixfaces of a cube, with the center of the cube at theCG (Fig. 1). The sensing axis of each accelerometeris along one of the diagonals of the cube face onwhich it lies, with opposing sensors using diago-nals that are crossed. (The diagonals will form aregular tetrahedron.)

We have found that a much more general geome-try is possible, with most of the advantages of theone in Reference 3. We found that the cube can bereplaced by an arbitrary parallelopiped. That means

ω

ω

ω

FY 98 6-23

We have developed a new type of Inertial Measurement Unit (IMU) in support of flight tests, basedon a set of linear accelerometers distributed inside the flight vehicle. This novel, gyroscope-freedesign overcomes the restriction for sensors at or near the body’s center of gravity (CG). The IMU iscapable of determining the kinematics of a rigid body with six degrees of freedom. We have developedthe mathematical model and are currently building the hardware for tests.

Carlos A. Avalle Defense Sciences Engineering DivisionElectronics Engineering

John I. CastorDefense and Nuclear Technologies

620 Avalle_qk 7/27/99 9:36 AM Page 6-23


that the six sensor locations can be any six pointsthat form a figure—a general octahedron—with acenter of symmetry. The center of symmetry doesnot have to be the CG. The six sensors would beplaced at the six vertices of the octahedron. Thesensing directions are parallel to the diagonals ofthe parallelogram faces on which the sensors lie.Diametrically opposed sensors choose alternatediagonals of their respective parallel faces.

The only simplicity of the cubical layout ofReference 3 that is not preserved is the ease of solv-ing the six equations for the components of andthe components of Acm. With the more generalarrangement a non-trivial system of six linear equa-tions in six unknowns must be solved at each time.Since the matrix does not vary, most of the work canbe done in advance.

The system of linear equations to be solved is thefollowing:

, (2)

where the right-hand side is a column vector formedby letting the index i run from 1 to 6, that is, overthe six sensors, and S and T are 6x3 matricesdefined by

. (3)

The solution may be written

(4)in terms of matrices M and N obtained by inverting[S T]. The special geometry is responsible for the

˙ ,

–

ω

ω ω

= [ ]= ⋅ × ×[ ]M

A N n R

i

cm i i i

a

a

S R n

T n

i i

i

,

= ×[ ]= [ ]

S T n R

Ai i i

cm

– ˙

[ ]

= ⋅ × ×[ ]ω

ω ωa

ω

centripetal acceleration term not entering the resultfor After is obtained from the first equation, itis integrated forward in time to give the current ω,which is used in the second equation to find theacceleration of the CG. The other aspects of thenavigation problem are carried out exactly asdescribed above for the case of gyroscopes.

Progress

To evaluate the performance of the six-accelerometer design, we initially developed acomputer model of the system. In the model, the testobject is defined in terms of its mass and inertia.Input linear and angular forces on the test objectcan be defined at multiple locations, and theseforces can be time varying. To evaluate the perfor-mance of the system, the model takes into accountsensor locations and orientations, sensitivity, accu-racy, integration times, and sampling periods. Themodel predicts velocities and acceleration timehistories relative to inertial space, at any number ofbody-fixed points, in any body-fixed direction.

Currently, we are completing hardware develop-ment and fabrication for a test flight. Our system isbased on Allied Signal QA-3100 inertial-gradeaccelerometers. These ultra-sensitive sensors arecapable of resolution down to <1 x 10–7 g, withfrequency response DC < f < 1 kHz.

Special signal conditioning was developed forour application. The accelerometer and signalconditioning hardware are shown in Figs. 2 and 3,in a photo, and a block diagram, respectively. Oneapplication required dynamic scale ranging toprevent high-frequency vibration and shock signalsdue to launch and flight vibration from saturating

ωω


Figure 1. Six-accelerometer IMU configuration. Figure 2. Photo of the accelerometer and signalconditioning hardware.

620 Avalle_qk 7/27/99 9:36 AM Page 6-24


the sensor’s output. These shock levels wereunknown, thus requiring laboratory and fieldexperiments to characterize and verify theresponse of sensors in these environments.

Initial attempts were made to isolate the sensorsmechanically. However, ultimately that approachwas supplanted by the electronic technique. Theseare true DC servo accelerometers, and as such, theyexhibit a small output bias that is nulled prior tobuffering and low-pass filtering. The accelerometersare also equipped with internal temperaturesensors. Acceleration and temperature signals aredigitized simultaneously, and a fourth-order temper-ature model is applied off-board for correction.

As presently configured for our flight test, theunit is set for a full scale range of ±0.5 g. A stan-dard 12-bit A/D converter provides a resolution of<±0.000250 g, which translates to differential veloci-ties on the order of 0.1 in./s and angular rates downto 0.25°/s for a duration on the order of 1 s.

In our flight test, acceleration and temperaturedata will be telemetered to ground stations andsignals processed off-board. Control or closed-loopnavigation applications could be realizable by on-board signal processing, the algorithms based onequations presented earlier.

Future Work

In the period of one year we have completed thedevelopment and are currently building the hard-ware for flight tests. The system has been optimizedto work in a zero-g environment, which would bedifficult to test in the laboratory. We are takingadvantage of an opportunity to test the unit in an up-coming flight test and are presently completing inte-gration of the system into the flight vehicle. Resultsof the test will not be known for several months, atwhich time a determination will be made regardingfurther development.

References

1. Regan, F. J., and S. M. Anandakrishnan (1993),Dynamics of Atmospheric Re-Entry, AIAA,Washington, D.C., Ch. 14.

2. Padgaonkar, A. J., K. W. Krieger, and A. I. King(1975), “Measurement of Angular Acceleration of aRigid Body Using Linear Accelerometers,” ASMEJournal of Applied Mechanics, 42, pp. 552-556.

3. Chen, J.-H., S.-C. Lee, and D. B. De Bra (1994),“Gyroscope Free Strapdown Inertial MeasurementUnit by Six Linear Accelerometers,” Journal ofGuidance and Control, 17, pp. 286-290.

FY 98 6-25

Accelerometer

Acceleration

Temperature

Excitation

To telemetry

Signal conditioning board

Externalpower

Voltageregulation

Biascompensation

BufferAnti-aliasfilter

Diff.amp.

Dynamicranging

Rangeoffset

Mux /ADC

Figure 3. Blockdiagram of theaccelerometer andsignal conditioningelectronics.

620 Avalle_qk 7/27/99 9:36 AM Page 6-25

iber-Based Phase-Shifting Diffraction Interferometerfor Measurement and Calibration of the Lick Adaptive Optics System


Introduction

This project was initiated to integrate a PSDI intothe adaptive optics (AO) system developed byLawrence Livermore National Laboratory (LLNL) foruse on the Shane telescope at Lick Observatory.Adding an interferometer to the AO system is usefulfor calibrating the control sensors, measuring theaberrations of the entire AO optical train, andmeasuring the influence functions of the individualactuators on the deformable mirror. A PSDI isparticularly well suited for this application. A PSDIoperates by using diffraction from a point-like aper-ture to generate a highly spherical wave that iscompared interferometrically to an aberrated spher-ical wavefront.1

Since the Lick AO system can be considered ablack box that relays an aberrated point imaged to acorrected, diffraction-limited point image, the refer-ence wave output by the PSDI can be fed withoutmodification into the AO system. Likewise, thecorrected output of the AO system can be fed with-out modification into the input of the PSDI. Thus, theonly aberrations measured by the PSDI will be thoseof the AO system. This provides an extremely accu-rate measurement of the optical properties of theAO system.

Usually, the input to the AO system is the imagecreated by the 3-m Shane telescope.2 Because ofatmospheric turbulence, this image is distorted andblurred. The AO system uses a fast moving tip-tiltmirror, which corrects for the blurring due to imagemovement, and a deformable mirror, which correctsfor the image distortion.

The tip-tilt mirror is placed in the expandingbeam, and the deformable mirror is placed in colli-mated light between two parabolic mirrors. The firstparabola collimates the input point image, and thesecond parabola focuses the corrected planar wave-front to create a corrected point image. This imageis then re-imaged by a scientific camera operating inthe infrared.

There are also six auxiliary optics in the system:two are dichroic mirrors, used for splitting off lightfor the sensors that control the tip-tilt anddeformable mirrors, and four are beam-steeringmirrors (Fig. 1).

All totalled, the image created by the AO systempasses through or reflects off of 11 optical elements.For this reason, it is not sufficient to simply replacethe deformable mirror with a flat mirror and measurethe aberrations right after the flat, as is currentlydone. This procedure accounts for all the aberrationsup to the flat, but leaves out the aberrations intro-duced beyond it. Using the PSDI to measure the

FY 98 6-27

An all-fiber based phase-shifting diffraction interferometer (PSDI) has been developed and inte-grated into the Lick Observatory adaptive optics system. Preliminary testing shows that the interfer-ometer has a single measurement accuracy of 18 nm RMS, and can achieve better than 6 nm RMSwith nine averages. The PSDI now needs to be incorporated into the control loop for the deformablemirror. The system then can be tested during an actual run on the Shane telescope. There are a fewengineering difficulties to be overcome, but their solutions are straight forward.

Eugene W. CampbellAdvanced Microtechnology ProgramLaser Programs

Jong R. AnLaser Engineering DivisionElectronics Engineering

625 Campbell_qk 7/27/99 9:38 AM Page 6-27


system aberrations allows the wavefront sensor to becalibrated so that it corrects the image in a way thataccounts for the aberrations of the entire optical train.

Progress

Design of the PSDI System

A difficulty in integrating a PSDI into the Lick AOsystem is that limited space is available for addinghardware. The PSDI systems developed at LLNLhave a footprint of approximately 3 ft × 3 ft. To makea system with a much smaller footprint, an entirelyfiber optic system was developed. This system occu-pies a footprint of 9 in. × 12 in., approximately 1/12the size of the discrete systems. The vertical dimen-sion also has been reduced, from 12 in. to 8 in. ACAD drawing of the system is shown in Fig. 2.

The all-fiber PSDI uses a 690-nm laser diode toprovide 10 mW of optical power with a coherencelength of approximately 3 m. This light is separatedinto two fibers by a variable beam-splitter that is setso that the beam intensities of each arm of the inter-ferometer are equal. One fiber goes to the phase-shifter in the reference wave arm of the interferome-ter, and the other fiber goes to a fiber spool that isused to equalize the optical path lengths betweenthe reference and test arms.

From there, the fibers go to polarization controllers(PLC) that are set so that the polarizations of the testand reference waves are identical at the camera. Uponleaving the PLCs, the fibers run to opposite ends ofthe AO system. The test fiber runs to the input point ofthe AO system, and the reference fiber runs to theoutput point of the AO system.

The phase-shifter is a commercially availabledevice that consists of 25 windings of fiber aroundan oval spool. The spool has piezoelectric (PZT)plates along the two long sides. The fiber is epoxiedto the PZT plates so that when a voltage is appliedto the plates, the fiber is stretched as the platesgrow in length. The fiber-based variable beam-splitter is also a commercial device, and is used tobalance the beam powers at the CCD camera toachieve maximum fringe contrast.

Once the test wave passes through the AOsystem, it converges onto the end of the referencefiber. It then reflects off the reference fiber tocombine with the diverging reference wavefrontdiffracting out of the end of the fiber. The end face ofthe reference optical fiber is super-polished (RMSroughness < 1 Å) to ensure that the reflected light isnot distorted by the shape of the fiber end face.

These two beams are then steered by a knife-edgemirror through an imaging lens and onto a CCDcamera. The knife-edge mirror passes half of the cone


Tip-tilt mirror

Collimating mirror

Deformable mirror

Focusing mirror

Blurred and aberrated image (from telescope)

Corrected image

(to camera)

Figure 1. Simplifiedrepresentation of theLick Observatory AOsystem.



of light diffracting from the reference fiber, andreflects half towards the camera. Since the numericalaperture of the beam exiting the AO system is verysmall (f# ≈ 27), it is not clipped by the knife-edge asit focuses onto the reference fiber. This opticalsystem that allows for the interference of the beams’test and reference waves is shown in Fig. 3.

Accuracy of the Initial System Test

The PSDI was mounted on the AO system bread-board and initial repeatability tests were performed. Itwas expected that air currents in the 5.8-m opticalpath and vibrations of the eight reflective optics wouldlead to noise in the measurement. However, since

FY 98 6-29

Fiber phase–shifter

Fiber polarizationcontroller

Optical pathcompensation spool

Shifter electronics

690–nm laser diode

Fiber variablebeam splitter

Figure 2. Front-enddesign of the PSDIsystem.

Optical fiber

Kodak ES1.0CCD camera

Fiber positioner

Knife edgemirror

Imaging lens

Figure 3. Opticalsystem. The referencewavefront diffractingfrom optical fibercombines with testwavefront from theAO system and inter-feres at the camera.



such noise is typically random, it is possible to makemultiple measurements and average them to reducethe noise. Unfortunately, feedback into the lasercaused instabilities in the source coherence, and itwas necessary to take measurements one at a time.Each measurement was analyzed on the spot, andapproximately one quarter of the measurements wererejected due to laser instability. A total of 25 measure-ments were kept. In the future, a Faraday-type opticalisolator will be used to prevent laser feedback.

The complete set of 25 measurements wasaveraged together to form a baseline measure-ment. Then each measurement was individuallycompared to the baseline. The typical differencebetween a single measurement and the baselinewas 18.6 nm RMS, with a standard deviation of9.1 nm RMS. Averaging groups of four measure-ments, and comparing them to the baselineyielded an error of 9.7 nm RMS, with a standarddeviation of 3.9 nm RMS.

As expected for N measurements of a system withrandom noise, the RMS error dropped as 1/ . Asgroups of nine and 16 were compared to the total set,

N

the noise dropped more quickly. This is because thesegroupings no longer appear like independentmeasurements when compared to a baseline formedwith only 25 measurements. The various groupingsare given in Table 1.

A histogram for the single measurement results isshown in Fig. 4, where each bin is 5 nm wide. Theminimum difference was 7.8 nm RMS, and the maxi-mum was 48 nm RMS. As can be seen, it is reason-able to treat the noise as random, and nine or moreaverages should yield a measurement with an RMSerror of less than 10 nm. To achieve an accuracy ofbetter than 1/100 of a wave in the visible region,approximately 16 averages would be required. At aprocessing time of 8 s/average, it will take approxi-mately 2 min to make a measurement accurate tobetter than 1/100 of a wave.

Wavefront Measurement at theDeformable Mirror

In all the measurements, the imaging lens wasfocused on the deformable mirror, which forms thepupil of the AO system. Therefore, the optimal defor-mations of the mirror needed to improve the wave-front can be calculated if the influence functions ofthe individual actuators are known.

A measurement of the wavefront at the deformablemirror is shown in Fig. 5. The peak-to-valley deviationof this wavefront is 1.65 µm, and the RMS deviation is133 nm. This wavefront is flat to within 1/6 of the632.8 nm measurement wavelength, which corre-sponds to a Strehl ratio of 0.36. The accuracy of thismeasurement is approximately 4 nm RMS. When the


0

1

2

3

4

5

6

7

8

9

10

More453525155Bin (nm RMS)

Freq

uen

cy (

coun

ts)

Figure 4. Histogram of the single measurement errors.

Figure 5. Wavefront of the AO system imaged at thedeformable mirror.

Table 1. Data from repeatability tests.

Number of Average RMS Standardaverages error (nm) deviation

1 18.6 9.084 9.66 3.919 5.23 —

16 3.58 —



wavefront has been freshly flattened using the phasediversity method, Strehl ratios of 0.45 have beenobtained. Interactive manipulations of the actuatorshave yielded Strehl ratios as high as 0.6.

It is expected that the PSDI can be used to flattenthe wavefront to obtain Strehl ratios better than 0.8.It should be noted that improving the Strehl ratiofrom 0.36 to 0.8 at 0.532 µm, only improves theStrehl ratio at 2.2 µm from 0.92 to 0.98. The realadvantage in using the PSDI to calibrate the systemis that it takes much less time than the currentmethod and is more reliable.

Influence Function of a Single Actuator

A measurement can also be made when a singleactuator is either pushed or pulled. The baselinemeasurement can then be subtracted from thismeasurement, and the difference yields the influencefunction for the particular actuator. Figure 6 shows

two measurements of the influence function ofmirror actuator #14 as it is pushed and pulled. Ascan be seen in both measurements, the influencefunction of the actuator is roughly Gaussian innature. However, there is a slight hexagonal shapeto the deflection because of the hexagonal layout ofthe actuators. Lineouts through the centers of thedeflections better display the magnitude of the wave-front change and are also shown in Fig. 6.

Note that this is not equal to the actual mirrordeformation. The wavefront deformation is doubledrelative to the mirror deformation because of thereflection, and there is also an obliquity factor dueto the light not being normally incident. The “pull”data is for a D/A voltage of –4.569 V on the actua-tor, and the “push” data is for D/A voltage of4.781 V. An image of the fringes in the pushing andpulling modes is shown in Fig. 7. Note that ~30 to60 tilt fringes have been introduced to overcomeproblem of multiple pass noise.

FY 98 6-31

300 400 500 600 700 300 400 500 600 700 800 700-0.50

Dif

fere

nce

(µm

)

0.00

0.50

1.00

1.50

-3.0

-2.0

-1.0

0.0

1.0

Pixel Pixel

Dif

fere

nce

(um

)

(a) (b)Pushing actuator #14 Pulling actuator #14

Figure 6. Measurements of influence function of mirror actuator as it is a) pushed and b) pulled.



Actuators at the center of the field and at theedge were also moved to verify the actuator locationmap. The images in Fig. 8 show the wavefrontdifference from the baseline as actuators #02 and#31 were pulled.

During the first year of this experiment, we accom-plished the following: 1) developed an all-fiber PSDIsystem of extremely small size; 2) qualified a fiber-opticphase-shifter to better than 1 nm accuracy; 3) wrotesoftware to control the PSDI and acquire data from aPC running Windows NT; 4) integrated the PSDI systemonto the AO bench in a way that least disturbed theexisting layout; and 5) used the PSDI system to makepreliminary measurements of the AO system.

Future Work

Development of the PSDI system brought to lightseveral problems that remain to be solved:

1. Laser source instability is limiting the accuracyof the PSDI system.

2. Light from the reference fiber is passing throughthe system, reflecting off the end of the inputfiber, passing back through the system, andcreating spurious noise fringes.

3. Both dichroics in the AO system must beremoved to use the PSDI because the dichroicsdo not pass sufficient energy at the PSDI lasersource wavelength.


Pushing actuator #14 Pulling actuator #14

(a) (b)

Figure 7. Image offringes in the a)pushing and b)pulling modes.

Pushing actuator #02 Pulling actuator #31Figure 8. Imagesshowing wavefrontdifference from base-line as actuatorswere pulled.



4. The PSDI system has yet to be used while theAO system is mounted on the Shane telescope.

5. Some of the more expensive equipment is onloan from the EUVL program and needs to bereplaced.

6. If the system is to become a permanent featureat Lick Observatory, it will have to be convertedto use a Sun workstation.

References

1. Sommargren, G. E. (1996), “Diffraction methodsraise interferometer accuracy,” Laser Focus World,pp. 61–71, August.

2. Olivier, S. S., and C. E. Max (1997), “First significantimage improvement from a sodium-layer laser guidestar adaptive optics system at Lick Observatory,”Lawrence Livermore National Laboratory, Livermore,California (UCRL-JC-128088).

FY 98 6-33


Selected EngineeringPublications



Aceves, S. M., and G. D. Berry (1998), “Onboard StorageAlternatives for Hydrogen Vehicles,” Energy andFuels, Vol. 12, No. 1, pp. 49–55.

Aceves, S. M., and G. D. Berry (1998), “Thermodynamicsof Insulated Pressure Vessels for Vehicular HydrogenStorage,” American Society of Mechanical EngineersJournal of Energy Resources Technology, Vol. 120,No. 2.

Aceves, S. M., and J. Martinez (1997), “Effects ofEvaporator Frosting on the Performance of Air-to-Air Heat Pump,” Proceedings of the AmericanSociety of Mechanical Engineers Advanced EnergySystems Division, M. L. Ramalingam, J. L. Lage, V.C. Mei, and J. N. Chapman, eds., November, AES-Vol. 37, pp. 357–363.

Aceves, S. M., and J. R. Smith (1997), “Hybrid andConventional Hydrogen Engines that Meet EZEVEmissions,” Paper 970290, Society of AutomotiveEngineers Transactions.

Aceves, S. M., C. Westbrook, W. Pitz, and J. R. Smith(1997), “Modeling of Homogeneous ChargeCompression Ignition (HCCI) of Methane,” PaperNo. 97-ICE-68, Proceedings of the 1997 AmericanSociety of Mechanical Engineers InternalCombustion Engine Fall Technical Conference, ICE-Vol. 29-3, pp. 85–90.

Aceves, S. M., and J. R. Smith (1998), “A DesiccantDehumidifier for Electric Vehicle Heating,” AmericanSociety of Mechanical Engineers Journal of EnergyResources Technology, Vol. 120, No. 2.

Aceves, S. M., G. D. Berry, and G. D. Rambach (1998),“Insulated Pressure Vessels for Hydrogen Storage onVehicles,” International Journal of Hydrogen Energy,Vol. 23, No. 7, pp. 583–591.

Aceves, S. M., H. Nakamura, G. M. Reistad, and J.Martinez-Frias (1998), “Optimization of a Class ofLatent Thermal Energy Storage Systems WithMultiple Phase Change Materials,” American Societyof Mechanical Engineers Journal of Solar EnergyEngineering, Vol. 120, No. 1, pp. 14–19.

Aceves, S. M., and B. T. Kornblum (1997), “ThermalAnalysis of Simulated Pantex Pit Storage,” AmericanSociety of Mechanical Engineers Proceedings of theThirty-Second National Heat Transfer Conference,August, Vol. 10, pp. 25–30.

Ackler, H. D., and Y.-M. Chiang (1997), “Model Experimenton Thermodynamic Stability of Retained IntergranularAmorphous Films,” Journal of the American CeramicSociety, July, Vol. 80, No. 7, pp. 1893–1896.

Alesso, H. P., C. E. Annese, and S. Murty (1997), “OnEstablishing Benchmark Standards for ParallelProcessing Monte Carlo Codes,” AmericanNuclear Society, Nuclear Crit ical i ty SafetyDivision, September.

Altenbach, T., and S. Brereton (1997), “Risk RankingMethodology for Radiological Events,” EFCOGSeventh Annual Safety Analysis Working Group,Oakland, California, June.

Avicola, K., J. A. Watson, B. V. Beeman, T. C. Kuklo, and J.R. Taylor (1998), “Design and Performance of the Tip-Tilt Subsystem for the Keck II Telescope and AdaptiveOptics System,” SPIE International Symposium onAstronomical Telescopes and Instrumentation, Kona,Hawaii, March, Vol. 3353.

Blaedel, K. (1998), “Ductile-Regime Grinding of BrittleMaterials,” Machining of Ceramics and Composites,Chapter 6, Marcel Dekker, New York, New York.

Brase, J. M., J. R. An, K. Avicola, B. V. Beeman, G. L.Berry, B. Johnston, C. E. Max, J. A. Watson, and K. E.Waltjen (1998), “Keck Wavefront Control System:Architecture and Initial Laboratory Test Results,”SPIE International Symposium on AstronomicalTelescopes and Instrumentation, Kona, Hawaii,March, Vol. 3353.

Brereton, S. J., G. Brumburgh, D. Becker, J. Pryatel, R.Wolfe, and J. Yatabe (1998), “Assuring Safety in theNational Ignition Facility,” EFCOG Eighth AnnualSafety Analysis Working Group Workshop, April.

Brown, N. W., J. Hassberger, E. Greenspan, and E. Elias(1997), “Proliferation Resistant Fission EnergySystems,” Global ‘97, October 5–10.

Brown, S. B., and F. P. Milanovich (1997), PatentDisclosure and Record of Invention: Number IL-10158, “Simple, Inexpensive, Efficient Coupling ofLight Emitting Diode (LED) Radiation into an OpticalFiber,” May 29.

Brummond, W., and G. Armantrout (1998), “PlutoniumImmobilization Project,” American Institute ofChemical Engineers Spring National Meeting, NewOrleans, Louisiana, March.

Bulmer, R. H., and G. H. Neilson (1997), “AlternativePoloidal Field Configurations for ITER,” Proceedingsof the Seventeenth IEEE/NPSS Symposium on FusionEngineering, San Diego, California, October 6–10.

Burnett, G. C., T. J. Gable, J. F. Holzrichter, and L. C. Ng(1997), “Voiced Excitation Functions Calculated fromMicropower Impulse Radar Information,” Paper4aSP4, One Hundred Thirty-Fourth Meeting of theAcoustic Society of America, San Diego, California,December 1–5.

Campbell, G. H., S. M. Foiles, H. Huang, D. A. Hughes,W. E. King, D. H. Lassila, D. J. Nikkel, T. D. de laRubia, J. Y. Shu, and V. P. Smyshlyaev (1998),“Multi-Scale Modeling of Polycrystal Plasticity: AWorkshop Report,” Materials Science andEngineering A, in press.

Y 98 PB-1


Canaan, R. E., J. E. Dec, R. M. Green, and D. T. Daly(1998), “The Influence of Fuel Volatility on the Liquid-Phase Fuel Penetration in a Heavy-Duty D. I. DieselEngine,” Paper No. 980510, Society of AutomotiveEngineers International Conference and Exposition,Detroit, Michigan, February.

Carrano, C. J., S. S. Olivier, J. M. Brase, B. A. Macintosh,and J. R. An (1998), “Phase Retrieval Methods forAdaptive Optics,” SPIE International Symposium onAstronomical Telescopes and Instrumentation, Kona,Hawaii, March 20–28, Vol. 353-111.

Carter, P. H. (1997), “HyperSoar: A Concept for GlobalReach and Space Access,” Strategic AvionicsTechnology Working Group NASA/Industry NationalMeeting, Seattle, Washington, October 21–23.

Carter, P. H., D. J. Pines, and L. vE Rudd (1998),“Approximate Performance of Periodic HypersonicCruise Trajectories for Global Reach,” accepted byAIAA Journal of Aircraft.

Carter, P. H., D. J. Pines, and L. vE Rudd (1998),“Approximate Performance of Periodic HypersonicCruise Trajectories for Global Reach,” AIAA 98-1644,Eighth AIAA International Spaceplanes andHypersonics Systems and Technologies Conference,Norfolk, Virginia, April 27–30.

Carter, P. H., F. Mitlitsky, A. H. Weisberg, J. C. Whitehead,and R. W. Humble (1998), “Design Trade Space for aMars Ascent Vehicle for a Mars Sample ReturnMission, IAA98-066,” Third IAA InternationalConference on Low-Cost Planetary Missions,Pasadena, California, April 27–May 1.

Cartland, H. E., and J. W. Hunter (1997), “The SHARPLight Gas Gun,” Proceedings of the Forty-ThirdInternational Instrumentation Symposium, Orlando,Florida, May 4–8, pp. 501–507.

Chang, J. J., E. P. Dragon, and I. L. Bass (1998), “315 WPulsed-Green Generation With a Diode-PumpedNd:YAG Laser,” Paper CPD2, Supplement toTechnical Digest, Conference on Lasers and Electro-Optics, San Francisco, California, May 3–8.

Chang, J. J., E. P. Dragon, C. A. Ebbers, and I. L. Bass(1998), “An Efficient Diode-Pumped Nd:YAG LaserWith 451 W of CW IR and 182 W of Pulsed GreenOutput,” Paper PDP-15, Supplement to TechnicalDigest, Advanced Solid-State Lasers Conference,Coeur d’Alene, Idaho, February 2–4.

Chen, Y. J., and G. Caporaso (1997), “Designs for aHigh Power Superconducting Delay Line,” 1997Pulsed Power Conference, Baltimore, Maryland,June 29–July 2.

Chen, Y. J., and G. J. Caporaso (1997), “A Novel Designfor a High Power Superconducting Delay Line,” 1997Particle Accelerator Conference, Vancouver, BC,Canada, May 12–16.

Chow, R., J. R. Taylor, Z. L. Wu, Y. Han, and T. Yang(1998), Absorptance Measurements of TransmissiveOptical Components by the Surface Thermal LensingTechnique,” Laser-Induced Damage in OpticalMaterials: 1997, G. J. Exarhos, A. H. Guenther, M. R.Kozlowski, and M. J. Soileau, eds., SPIE Proceedings,Vol. 3244, pp. 376–385.

Chow, T. S., P. J. Harwood, and S. J. DeTeresa (1997),“Characterization of Cellular Silicone for WeaponSystem Applications,” Twenty-First Compatibility,Aging, and Stockpile Stewardship Conference, KansasCity, Kansas, September.

De Groot, A. J., and D. Harris (1998), “ComputationallyEfficient, Robust Algorithm for Matched FieldProcessing,” Circuits, Systems, and SignalProcessing, April, Vol. 17, No. 2, pp. 165–193.

De Groot, A. J., R. J. Sherwood, D. C. Badders, and C. G.Hoover (1997), “Parallel Contact Algorithms forExplicit Finite Element Analysis (DYNA3D),”Proceedings of the Fourth U. S. National Congress onComputational Mechanics, San Francisco, California,August 6–8.

De Groot, W. A., L. A. Arrington, J. F. McElroy, F. Mitlitsky,A. H. Weisberg, P. H. Carter II, B. Myers, and B. D.Reed (1997), “Electrolysis Propulsion forSpacecraft,” AIAA Joint Propulsion Conference,Seattle, Washington.

Dec, J. E., and R. E. Canaan (1998), “PLIF Imaging of NOFormation in a D. I. Diesel Engine,” Paper No.980147, Society of Automotive EngineersInternational Conference and Exposition, Detroit,Michigan, February.

Decker, J. Y., A. Fernandez, and D. W. Sweeney (1997),“Generation of Subquarter-Micron Resist StructuresUsing Optical Interference Lithography and ImageReversal,” Journal of Vacuum Science Technology,November/December, B 15 No. 6.

Dehghani, M. (1997), “Numerical Simulation ofConventional Spin Forming,” Fourth U.S. NationalCongress on Computational Mechanics, SanFrancisco, California, August 6–8, p. 250.

DelGrande, N. (1997), “Heat Capacity Mapping forQuality Evaluation of Weapons Materials andStructures,” Twenty-First Compatibility, Aging, andStockpile Stewardship Conference, Kansas City,Kansas, August.

Demos, S., M. Yan, B. Woods, M. Staggs, Z. L. Wu, H. B.Radouski, and J. De Yoreo (1998), “Temperature andSpectral Investigation of Bulk KDP Below DamageUsing 355 nm Laser Irridiation,” Laser-InducedDamage in Optical Materials: 1997, G. J. Exarhos, A.H. Guenther, M. R. Kozlowski, and M. J. Soileau, eds.,SPIE Proceedings, Vol. 3244, pp. 223–227.

Engineering Research Development and Technology PB-2


Domansky, K., D. L. Baldwin, J. W. Grate, T. B. Hall, J. Li,M. Josowicz, and J. Janata (1998), “Developmentand Calibration of Field-Effect Transistor-BasedSensor Array for Measurement of Hydrogen andAmmonia Gas Mixtures in Humid Air,” AnalyticalChemistry, February 1, Vol. 70, No. 3, pp. 473–481.

Edmunds, T. (1997), Nondifferentiable and Two-LevelMathematical Programming, (by Kiyotaka Shimizu, YoIshizuka, and Jonathan F. Bard, Kluwer), Section15.4, pp. 371–380.

Erbert, G., K. Kanz, and I. Bass (1997), “Lamp PumpedNd:YAG Laser Oscillator With 170 W Green Output”Paper PD2.6, IEEE Lasers and Electro-Optics SocietyAnnual Conference, San Francisco, California,November 10–13.

Fabbricator, P., S. Farinon, R. Musenich, C. Priano, T. G.O’Connor, R. A. Bell, W. Burgess, W. Craddock, R.Penco, and P. Valente (1997), “The SuperconductingSolenoid for The BABAR Experiment at PEP-II inSLAC,” Fifteenth International Conference on MagnetTechnology, Beijing, China, October.

Feit, M. D., J. Campbell, F. Genin, A. Salleo, D. Faux, A. M.Rubenchik, R. Riddle, M. R. Kozlowski, and J.Yoshiyama (1997), “Analysis Of Laser-InducedSurface Cracks in Silica at 355 nm,” Twenty-NinthAnnual Symposium on Optical Materials for HighPower Lasers, Boulder, Colorado, October 6–8, (toappear in SPIE).

Ferguson, J. (1997), “Simultaneous Monitoring of pH, CO2and O2, Using an Optical Imaging Fiber,” AnalyticaChimica Acta, Vol. 340, pp. 123–131.

Fernandez A., J. Y. Decker, S. M. Herman, D. W. Phillion,and D. W. Sweeney (1997), “Methods for FabricatingArrays of Holes Using Interference Lithography,”Journal of Vacuum Science Technology,November/December, B 15 No. 6.

Folta, J. A., C. Yu, W. J. Benett, and D. Ciarlo (1997), “AMiniature DNA Based Analytical Instrument,” Journalof Analytical Chemistry, May.

Fulkerson, E. S., D. C. Norman, and R. Booth (1997),“Driving Pockel Cells Using Avalanche TransisterPulsers,” Eleventh IEEE International Pulsed PowerConference, June 29–July 3.

Gable, T. J., G. C. Burnett, J. F. Holzrichter, L. C. Ng, andW. A. Lea (1997), “Comparison of ConventionalAcoustic and MIR Radar/Acoustic Processing ofSpeech Signals,” One Hundred Thirty-Fourth Meetingof the Acoustic Society of America, December 1–5,San Diego, California, p. 139.

Garcia, M. (1997), “Designing Planar MagnetronCathodes: Analysis and Experiment,” IEEEConference Record - Abstracts, IEEE InternationalConference on Plasma Science, May 19–22, SanDiego, California.

Glosup, J., and M. Axelrod (1997), “Systematic ErrorRevisited,” Annual Meeting of the AmericanStatistical Association, 1996 Proceedings for theSection on Physical and Engineering Sciences,pp. 81–86.

Goosman, D., G. Avara, L. Steinmetz, C. Lai, and S. Perry(1997), “Manybeam Velocimeter for Fast Surfaces,SPIE Proceedings of the Twenty-Second InternationalCongress on High-Speed Photography and Photonics,Vol. 2869, pp. 1070–1079.

Groves, S. E., S. J. DeTeresa, and R. J. Sanchez (1998),“Accelerated Durability and Long-Term Testing ofHigh-Temperature Polymer Composites,” ASTMSymposium on Time-Dependent and NonlinearEffects in Polymers and Composites, Atlanta,Georgia, May 4–5.

Hakins, J., and W. Jandeska (1998), “Power Flow andDie Filling Studies Using Computed Tomography,”Ninty-Eighth International Conference on PowerMetallurgy and Particulate Materials, Las Vegas,Nevada, May 5–June 4.

Han, Y., Q. Zhao, Z. L. Wu, and K. Mocur (1998), “Near-Field Detection of Laser-Induced Thermal Waves inOptical Materials,” Laser-Induced Damage in OpticalMaterials: 1997, G. J. Exarhos, A. H. Guenther, M. R.Kozlowski, and M. J. Soileau, eds., SPIE Proceedings,Vol. 3244, pp. 257–267.

Haney, S. W., L. D. Pearlstein, R. H. Bulmer, and J. P.Freidberg (1997), “Vertical Stability Analysis ofTokamaks Using a Variational Procedure,” PlasmaPhysics Reports, September, Michael Ioffe SpecialIssue #9.

Hartemann, F. V., A. L. Troha, G. P. LeSage, C. V. Bennett,B. H. Kolner, and N. C. Luhmann, Jr. (1997),“Ultrahigh Intensity Compton Scattering Focussed X-Ray Source,” IEEE Conference Record - Abstracts.1997 IEEE International Conference on PlasmaScience, San Diego, California, May 19–22, p. 271.

Hassberger, J. A., R. N. Schock, and T. H. Isaacs (1997),“Prospects of and Requirements for Nuclear Power asa Contributor Toward Managing Greenhouse Gases,”International Conference on Environment andNuclear Energy, October.

Havstad, M. A., and C. Dingus (1997), “RadiantTransmittance of Cerium Doped Quartz from 300 to1270 K,” Proceedings of the Thirty-Second NationalHeat Transfer Conference, Baltimore, Maryland,August 8–12, ASME HTD-Vol. 345, pp. 73–80.

Holdener, F. R. (1997), “Beam Control and DiagnosticFunctions in the NIF Transport Spatial Filter,” Solid-State Lasers for Applications to Inertial ConfinementFusion, Second Annual International Conference,Paris, France.

FY 98 PB-3


Holzrichter, J. F., G. C. Burnett, L. C. Ng, and W. A. Lea(1998), “Speech Articulator Measurements UsingLow Power EM-Wave Sensors,” Journal of theAcoustic Society of America, January, Vol. 103 No. 1,pp. 622–625.

Holzrichter, J. F., W. A. Lea, L. C. Ng, and G. C. Burnett(1997), “New Sensors Offer Breakthroughs in VoiceInterfacing,” AVIOS Conference, San Francisco,California, July.

Honea, E. C., R. J. Beach, S. B. Sutton, J. A. Speth, S. C.Mitchell, J. A. Skidmore, M. A. Emanuel, and S. A.Payne (1997), “115-W Tm:YAG Diode-Pumped Solid-State Laser,” IEEE Journal of Quantum Electronics,September, Vol. 33, No. 9, pp. 1592–1600.

Hoover, C. G., A. J. De Groot, D. C. Badders, and R. J.Sherwood (1997), “Parallel Algorithms for ExplicitFinite Element Analysis (DYNA3D),” Proceedings ofthe Fourth U. S. National Congress on ComputationalMechanics, San Francisco, California, August 6–8.

Hossain, Q. A. (1997), “Why a Performance Goal-BasedSeismic Method?” Proceedings of ASCE StructuresCongress XV, Vol. 1, April.

Houck, T. L., G. J. Caporaso, C. C. Shang, S. E. Sampayan,N. E. Molau, and M. L. Krogh, (1997), “Measured andTheoretical Characterization of the RF Properties ofStacked, High-Gradient Insulator Material,” 1997Particle Accelerator Conference, Vancouver, BC,Canada, May 12–16.

Hsiao, M. C., B. M. Penetrante, B. T. Merritt, G. E. Vogtlin,and P. H. Wallman (1997), “Reduction of NO2 in N2 byNon-Thermal Plasmas,” Journal of AdvancedOxidation Technologies, Vol. 2, pp. 283–285.

Hsiao, M. C., B. T. Merritt, B. M. Penetrante, G. E. Vogtlin,and P. H. Wallman (1997), “Effect of Gas Temperatureon Pulsed Corona Discharge Processing of Acetone,Benzene and Ethylene,” Journal of AdvancedOxidation Technologies, Vol. 2, pp. 306–311.

Huang, S. T., D. A. Lappa, and T. Chiao (1997), “Use of aComputer-Assisted Administrative Control toEnhance Criticality in LLNL for Fissile MaterialDisposition Operations,” Nuclear CriticalityTechnology Safety Project Annual Workshop,Gaithersburg, Maryland, May 5–9.

Huang, S., D. Lappa, T. Chiao, C. Parrish, R. Carlson,J. Lewis, D. Shikany, and H. Woo (1997), “Real-Time Software Use in Nuclear Materials HandlingCriticality Safety Control,” EFCOG SeventhAnnual Safety Analysis Working Group, Oakland,California, June.

Hurst, P. A (1998), “Optic Assembly and Alignment forthe National Ignition Faciltiy Project,” PhotonicsWest, January.

Johnson, Gary W. (1997), “LabVIEW GraphicalProgramming,” 2nd ed., McGraw-Hill, New York,New York.

Kennedy, K., B. Harteneck, G. Millos, M. Benapfl, K. King,and R. Kirby (1997), “TiN Coating of the PEP-II Low-Energy Ring Aluminum Arc Vacuum Chambers,”Twenty-Fifth Annual American Vacuum Symposiumand Exhibition, February 16.

Khanaka, G. H., and J. Yee (1997), “Semiconductors forRoom-Temperature, High-Energy RadiationDetection,” Arms Control and NonproliferationTechnologies, U.S. Department of Energy, Office ofNonproliferation and National Security, pp. 12–13.

Kim, J., J. Simon, S. Saffer, and C.-J. Kim (1998),“Mercury Contact Micromechanical Relays,”Proceedings of the Fourty-Sixth AnnualInternational Relay Conference, Oak Brook, Illinois,April, pp. 19-1–19-8.

Kondow, M. , T. Kitatani, S. Nakatsuka, M. C. Larson, K.Nakahara, Y. Yazawa, M. Okai, and K. Uomi (1997),“GaInNAs: A Novel Material for Long-WavelengthSemiconductor Lasers,” IEEE Journal of SelectedTopics in Quantum Electronics, June, Vol. 3, No. 3,pp. 719–730.

Kuzmenko, P. J., and D. Ciarlo (1998), “Improving theOptical Performance of Etched Silicon Gratings,”SPIE International Symposium on AstronomicalTelescopes and Instrumentation, InfraredAstronomical Instrumentation, March.

Kyle, K. R., and S. B. Brown (1997), “Fiber Optic RamanProbe for In Situ Chemical Characterization of theHanford Underground Storage Tanks,” Proceedingsof the Electrochemical Society, Chemical andBiological Sensors and Analytical ElectrochemicalMethods, Vol. 97-17.

Kyle, K. R., and S. B. Brown (1997), “Fiber Optic RamanProbe for the In Situ Cone Penetrometer ChemicalCharacterization of the Hanford Underground StorageTanks,” American Chemical Society National Meeting,Las Vegas, Nevada, September 7–11.

Kyle, K. R., and S. B. Brown (1997), “Fiber Optic RamanSpectroscopy for In Situ Waste TankCharacterization,” Thirty-Seventh ORNL-DOEConference on Analytical Chemistry in EnergyTechnology, Gatlinburg, Tennessee, October 8.

Kyle, K. R., and S. B. Brown (1998), “Fiber Optic RamanSpectroscopy Probe,” Department of ChemistryInorganic Seminar, University of California, SantaBarbara, California, March 11.

Larson, M. C., M. Kondow, T. Kitatani, K. Tamura, Y.Yazawa, and M. Okai (1997), “Photopumped Lasingat 1.25 mm of GaInNAs/GaAs Multiple Quantum WellVertical-Cavity Surface-Emitting Lasers,” IEEEPhotonics Technology Letters, December, Vol. 9, No.12, pp. 1549–1551.



Larson, M. C., M. Kondow, T. Kitatani, K. Nakahara, K.Tamura, H. Inoue, and K. Uomi (1998),“GaInNAs/GaAs Long-Wavelength Vertical-CavitySurface-Emitting Laser Diodes,” IEEE PhotonicsTechnology Letters, February, Vol. 10, No. 2, pp. 188–190.

Larson, M. C., M. Kondow, T. Kitatani, Y. Yazawa, and M.Okai (1997), “Room Temperature Continuous-WavePhotopumped Operation of 1.22-mm GaInNAs/GaAsSingle Quantum Well Vertical-Cavity Surface-Emitting Laser,” Electronics Letters, May, Vol. 33,No. 11, pp. 959–960.

Lavietes, A. D., J. H. McQuaid, and T. J. Paulus (1997),“Cadmium Zinc Telluride Detector System for NuclearMaterial Assay,” Proceedings of the Thirty-EighthAnnual Meeting, Institute of Nuclear MaterialManagement, July.

Lee, C. L., and Lee, C. T. (1997), “A Higher Order Methodof Multiple Scales,” Journal of Sound and Vibration,Vol. 202, No. 2, pp. 284–287.

Lee, J. D. (1997), “Criticality Safety Challenges in theNext Decade,” Proceedings of the American NuclearSociety, September 7–11.

LeSage, G. P., C. V. Bennett, W. E. White, E. C. Landahl,L. L. Laurent, N. C. Luhmann, Jr., F. V Hartemann,C. H. Ho, W. K. Lau, and T. T. Yang (1998), “A HighBrightness, X-Band Photoinjector for the Productionof Coherent Synchrotron Radiation,” APSConference on Physics of Plasmas, November15–21, Vol. 5, No. 5.

LeSage, G. P., C. V. Bennett, L. L. Laurent, J. A. Van Meter,V. Dinh, A. L. Troha, B. H. Kolner, F. V. Hartemann,and N. C. Luhmann, Jr. (1997), “Cold and High-PowerTests of a Multibunch X-Band Photoinjector,” IEEEInternational Conference on Plasma Science, SanDiego, California, May 19–22, p. 130.

Lesuer, D. R., C. K. Syn, O. D. Sherby, and J. Wadsworth(1998), “Laminated Metal Composites - Fracture andBallistic Impact Behavior,” Advanced Materials andProcessing - PRICM 3, O. P. Aurora et al., eds.

Lesuer, D. R., C. K. Syn, O. D. Sherby, D. K. Kim, and W. D.Whittenberger (1997), “Mechanical Behavior ofUltrahigh Strength, Ultrahigh Carbon Steel Wire andRod,” Thermo-Mechanical Processing and MechanicalProperties of Hypereutectoid Steels and Cast Irons,D. R. Lesuer, C. K. Syn and O. D. Sherby, eds., TMS,Warrendale, Pennsylvania, pp. 175–188.

Lesuer, D. R., R. Glaser, and C. K. Syn (1998), “TheEvolution of Grain Size Distribution DuringDeformation of Superplastic Materials,”Superplasticity and Superplastic Forming - 1998, A.K. Ghosh and T. Biehler, eds., TMS, Warrendale,Pennsylvania, pp. 33–42.

Lesuer, D. R., R. Glaser, and C. Syn (1998), “The Evolutionof Grain Size Distribution and Its Influence on theBehavior of Superplastic Materials,” One HundredTwenty-Seventh Annual TMS Meeting, February.

Li, E., and M. Dehghani (1997), “Constitutive ModelDevelopment for U-6 Niobium Using Modified Tensileand Torsional Test Data,” Fourth U.S. NationalCongress on Computational Mechanics, SanFrancisco, California, August 6–8, p. 315.

Liberman, V., V. Malba, and A. F. Bernhardt (1997),“Integration of Vapor Deposited Polyimide into aMultichip Module Packaging Process,” IEEETransactions Composites, Hybrids, ManufacturingTechnology, Vol. 20, pp. 13–16.

Liberman, V., V. Malba, and A. F. Bernhardt (1997), “VaporDeposition of Polyimide: Monomer Segregation at thePolyimide/Si (100) Interface,” Thin Solid Films, Vol.305, pp. 26–29.

Logan, C. (1997), “Advancing Technologies andApplications in Nondestructive Evaluation,” Scienceand Technology, December, pp. 4–11.

Malba, V., A. Conder, and A. F. Bernhardt (1997), “AStacked-DRAM Solid State Recorder Using a NovelLaser 3-D Interconnect Process,” Proceedings of theThird International 3-D Packaging Workshop,Philadelphia, Pennsylvania, October.

Malba, V., V. Liberman, and A. F. Bernhardt (1997),“Fabrication of DRAM Stacks Using 3-D Laser DirectWrite Patterning of Positive ElectrophoreticPhotoresist,” International Journal of Microcircuitsand Electronic Packaging, Vol. 20, pp. 371–375.

Martz, H. (1997), “Gamma-Ray Scanner Systems for NDAssay of Heterogeneous Materials,” Symposium onInternational Safeguards, Vienna, Austria, October.

Martz, H. (1997), “The Role of NDE in Life CycleManagement,” Third Annual Symposium on Frontiersof Engineering, August.

Matthews, S. M., T. E. Cowan, and H. S. Peters (1997),“Irradiation of Potting Compound from FluorescentLight Ballasts for Treatment of PCB, EPRI MunicipalWater and Wastewater Program,” EnvironmentalApplications of Advanced Oxidation Technologies,Electric Power Research Institute Industrial andAgricultural Technologies and Services, September.

McCallen, R. C., B. T. Kornblum, and W. Kollman (1997),“Large-Eddy Simulation in Complex Domains Usingthe Finite Element Method, FEDSM97-3496,”Proceedings of the American Society of MechanicalEngineers Fluids Engineering Summer Meeting,Vancouver, Canada, BC, June 22–26.

FY 98 PB-5


McCallen, R. C., F. Browand, A. Leonard, and W. Rutledge(1997), “Systematic Approach to Analyzing andReducing Aerodynamic Drag of Heavy Vehicles,”Annual Automotive Technology DevelopmentCustomers’ Coordination Meeting, Dearborn,Michigan, October 27–30.

McKoon, R. H. (1997), “Characterization of Electron BeamMelted Uranium - 6% Niobium Ingots,” Proceedingsof the Conference on Electron Beam Melting andRefining - State of the Art, October 5–7, pp. 182–195.

Mill, T., M. Su, C. C. D. Yoa, S. M. Matthews, and F. T. S.Wang. (1997), “E-Beam Treatment ofTrichloroethylene-Air Mixtures: Products and Rates,”Radiation Physics and Chemistry, September, Vol. 50,No. 3.

Miller, W. (1977), “Improved Axisymmetric View FactorCalculations for FACET,” Tri-Lab EngineeringConference on Modeling and Simulation, November.

Miller, W., and L. Parietti (1997), “Thermal Analysis ofthe NIF Final Optics Assembly - Part 2,” Tri-LabEngineering Conference on Modeling andSimulation, November.

Murphy, K. D., and C. L. Lee (1998), “The 1:1 InternallyResonant Response of a Cantilever Beam Attached toa Rotating Body,” Journal of Sound and Vibration, Vol.211, No. 2, pp. 179–194.

Nakahara, K., M. Kondow, T. Kitatani, M. C. Larson, and K.Uomi (1998), “1.3-mm Continuous-Wave LasingOperation in GaInNAs Quantum-Well Lasers,” IEEEPhotonics Technology Letters, April, Vol. 10, No. 4,pp. 487–488.

Nieh, T. G., R. O. Kaibyshev, and D. R. Lesuer (1998),“Superplasticity in a Conventional Coarse-Grained6061 Al and the Liquid Phase Effect,” Superplasticityand Superplastic Forming - 1998, A. K. Ghosh and T.Biehler, eds., TMS, Warrendale, Pennsylvania, pp.137–144.

Northrup, M. A., B. Benett, D. Hadley, P. Landre, S. Lehew,J. Richards, and P. Stratton (1998), “A MiniatureAnalytical Instrument for Nucleic Acids Based onMicromachined Silicon Reaction Chambers,”Analytical Chemistry, March, Vol. 70, No. 5, pp.918–922.

Northrup, M. A., B. Benett, D. Hadley, P. Landre, S. Lehew,J. Richards, and P. Stratton (1997), “Aligned WaferBonding for Microfluidic Devices,” Proceedings of theElectrochemical Society International Conference,Paris, France, September.

Page, R. H., R. A. Bartels, R. J. Beach, S. B. Sutton, L. H.Furu, and J. E. LaSala (1997), “1-Watt Composite-Slab Er:YAG Laser,” OSA TOPS - Advanced SolidState Lasers, Vol. 10.

Papados, P. P., R. R. Namburu, C. G. Hoover and A. J. DeGroot (1997), “Scalability Studies for Soil-StructureInteraction Problems,” Proceedings of the Fourth U.S. National Congress on Computational Mechanics,San Francisco, California, August 6–8.

Parietti, L., and W. Miller (1997), “Thermal Analysis ofthe NIF Final Optics Assembly - Part 1,” Tri-LabEngineering Conference on Modeling andSimulation, November.

Penetrante, B. M., G. E. Vogtlin, B. T. Merritt, and R. M.Brusasco (1997), “Plasma Technology for Tail PipeReduction of NOx in Diesel Exhaust,” Proceedings ofthe 1997 South Coast Air Quality ManagementDistrict Symposium on Air Pollution Health Impacts:Recent Findings, Implications, Dieselization andPolicy Initiatives, Diamond Bar, California, November.

Penetrante, B. M., M. C. Hsiao, B. T. Merritt, and G. E.Vogtlin (1997), “Fundamental Limits on Gas-PhaseChemical Reduction of NOx in a Plasma,” Proceedingsof the 1997 Diesel Engine Emissions ReductionWorkshop, San Diego, California, July.

Penetrante, B. M., M. C. Hsiao, B. T. Merritt, and G. E.Vogtlin (1997), “Multi-Stage Selective CatalyticReduction of NOx in Lean-Burn Engine Exhaust,”Proceedings of the 1997 Diesel Engine EmissionsReduction Workshop, San Diego, California, July.

Penetrante, B. M., M. C. Hsiao, B. T. Merritt, and G. E.Vogtlin (1997), “Fundamental Limits on NOxReduction by Plasma,” Proceedings of the 1997Society of Automotive Engineers International SpringFuels and Lubricants Meeting and Exposition,Detroit, Michigan, May 5–8.

Penetrante, B. M., M. C. Hsiao, B. T. Merritt, G. E. Vogtlin,C. Z. Wan, G. W. Rice, and K. E. Voss (1997),“Plasma-Assisted Heterogeneous Catalysis for NOxReduction in Lean-Burn Engine Exhaust,”Proceedings of the 1997 Diesel Engine EmissionsReduction Workshop, San Diego, California, July.

Penetrante, B. M., M. C. Hsiao, J. N. Bardsley, B. T.Merritt, G. E. Vogtlin, A. Kuthi, C. P. Burkhart, and J.R. Bayless (1997), “Identification of Mechanisms forDecomposition of Air Pollutants by Non-ThermalPlasma Processing,” Plasma Sources Science andTechnology, Vol. 6, pp. 251–259.

Penetrante, B. M., M. C. Hsiao, J. N. Bardsley, B. T.Merritt, G. E. Vogtlin, A. Kuthi, C. P. Burkhart, and J.R. Bayless (1997), “Decomposition of MethyleneChloride by Electron Beam and Pulsed CoronaProcessing,” Physics Letters A 235, pp. 76–82.

Penetrante, B. M., M. C. Hsiao, J. N. Bardsley, B. T.Merritt, G. E. Vogtlin, A. Kuthi, C. P. Burkhart, and J.R. Bayless (1997), “Comparison of Pulsed Coronaand Electron Beam Processing of Hazardous AirPollutants,” Journal of Advanced OxidationTechnologies, Vol. 2, pp. 299–305.



Penetrante, B. M., W. J. Pitz, M. C. Hsiao, B. T. Merritt,and G. E. Vogtlin (1997), “Effect of Hydrocarbons onPlasma Treatment of NOx,” Proceedings of the 1997Diesel Engine Emissions Reduction Workshop, SanDiego, California, July.

Piscotty, M. (1997), “In-Process EDM Truing to GenerateComplex Contours on Metal-Bond, SuperabrasiveGrinding Wheels for Precision Grinding StructuralCeramics,” International Conference on PrecisionEngineering, November.

Piscotty, M., T. Saito, and P. Davis (1997), “In-ProcessEDM Truing to Generate Complex Contours on Metal-Bond, Super-Abrasive Grinding Wheels for PrecisionGrinding Structural Ceramics” Amercian Society forPrecision Engineering, Twelvth Annual Meeting,October 5.

Piscotty, M., T. Saito, and P. Davis (1997), “Use of In-Process EDM Truing to Generate Complex Contourson Metal-Bond, Super-Abrasive Grinding Wheels forPrecision Grinding Structural Ceramics,”International Conference on Precision Engineering,Taipei, Taiwan November 18–20.

Poole, B. R., G. J. Caporaso, and W. C. Ng (1997), “WakeProperties of a Stripline Beam Kicker,” 1997 ParticleAccelerator Conference, Vancouver, BC, Canada, May12–16.

Rainer, F. (1998), “Mapping and Inspection of Damage andArtifacts in Large-Scale Optics, Laser-InducedDamage in Optical Materials: 1997, G. J. Exarhos, A.H. Guenther, M. R. Kozlowski, and M. J. Soileau, eds.,SPIE Proceedings, Vol. 3244, pp. 272–281.

Rasty, J., N. Dutta, M. Dehghani, and M. Rassaian (1997),“Residual Stresses Induced by Co-Drawing of CircularRods,” Proceedings of the Fourth U.S. NationalCongress on Computational Mechanics, SanFrancisco, California, August, p. 312.

Ratowsky, R. P., L. Yang, R. J. Deri, J. S. Kallman, and G.Trott (1997), “Laser Diode to Single-Mode Fiber BallLens Coupling Efficiency: Full-Wave Calculation andMeasurements,” Applied Optics 2005, pp.3435–3438.

Ratowsky, R. P., S. Dijaili, J. S. Kallman, M. D. Feit, J.Walker, W. Goward, and B. B. Afeyan (1998),“Modeling a Distributed Spatial Filter Low-NoiseSemiconductor Optical Amplifier,” IntegratedPhotonics Research, Optical Society of AmericaTechnical Digest Series, Vol. 4, pp. 152–154.

Rudd, L. vE., D. J. Pines, and P. H. Carter (1998),“Improved Performance of Sub-Optimal PeriodicHypersonic Cruise Trajectories for Long Range,”AIAA Journal of Aircraft, May.

Rudd, L. vE., D. J. Pines, and P. H. Carter (1998),“Improved Performance of Sub-Optimal PeriodicHypersonic Cruise Trajectories for Long Range,”Eighth AIAA International Spaceplanes andHypersonics Systems and Technologies Conference,Norfolk, Virginia, April 27–30.

Rudd, L. vE., D. J. Pines, and P. H. Carter (1998),“Optimal Parameterized Damped PeriodicHypersonic Cruise Trajectories,” AIAA Journal ofAircraft, in press.

Ruiz, R., and W. McNab (1998), “Engineering Design andTesting of a Ground Water Remediation System UsingElectrolytically Generated Hydrogen with a PalladiumCatalyst for Dehalogenating ChlorinatedHydrocarbons,” Proceedings Part A-Technical/Professional Papers, 1998 MexicanAmerican Engineering Society National Symposium,San Diego, California, January 14–17.

Runkel, M., J. DeYoreo, W. Sell, and D. Milam (1998),“Laser Conditioning Study of KDP on the OpticalSciences Laser Using Large Area Beams,” Laser-Induced Damage in Optical Materials: 1997, G. J.Exarhos, A. H. Guenther, M. R. Kozlowski, and M.J. Soileau, eds., SPIE Proceedings, Vol. 3244, pp.51–63.

Runkel, M., M. Yan, J. DeYoreo, and N. Zaitseva (1998),“The Effect of Impurities and Stress on the DamageDistributions of Rapidly Grown KDP Crystals,” Laser-Induced Damage in Optical Materials: 1997, G. J.Exarhos, A. H. Guenther, M. R. Kozlowski, and M. J.Soileau, eds., SPIE Proceedings, Vol. 3244, pp.211–223.

Russell, E. W., and D. A. Lappa (1997), “Validation andVerification of Finite Element Method Codes Used forAnalyses of Pit Storage/Transport Packages,” SecondBiennial Tri-Laboratory Engineering Conference onModeling and Simulation, Santa Fe, New Mexico,November 12–14.

Sahai, V. (1997), “Influence of Bulk Convection onFreckle Formation in Castings,” Proceedings of theFourth International Special Emphasis Symposiumon Superalloys 718, 625, 706, and Derivatives,TMS, June.

Saito, T. (1997), “Policy and the Photon: A FederalBalancing Act,” Optical Engineering Reports, July,Vol. 163, p. 2.

Saito, T. (1997), “Policy and the Photon: Time forAttention at the National Level,” Optical EngineeringReports, May, Vol. 161, p. 2.

Salleo, A., F. Y. Genin, J. Yoshiyama, C. J. Stolz, and M. R.Kozlowski (1998), “Laser-Induced Damage of FusedSilica at 355 nm Initiated at Scratches,” Laser-Induced Damage in Optical Materials: 1997, G. J.Exarhos, A. H. Guenther, M. R. Kozlowski, and M. J.Soileau, eds., SPIE Proceedings, Vol. 3244, pp.341–347.

FY 98 PB-7


Sargis, P. D., B. D. Henderer, and M. E. Lowry (1997),“10-Gb/s Subcarrier Multiplexed Transmission Over490 km of Ordinary Single-Mode Fiber WithoutDispersion Compensation,” IEEE PhotonicsTechnology Letters, December, Vol. 9, No. 12, pp.1658–1660.

Schwartz, A. J., D. H. Lassila, and M. M. LeBlanc (1998),“The Effects of Tungsten Addition on the Microtextureand Mechanical Behavior of Tantalum Plate,”Materials Science and Engineering A, A244, pp.178–190.

Scott, R. P., C. V. Bennett, and B. H. Kolner (1997), “AMand High-Harmonic FM Laser Mode Locking,”Applied Optics, August, Vol. 36, No. 24, pp.5908–5912.

Shang, C. C., Y.-J. Chen, T. L. Houck, G. J. Caporaso, W. C.Ng, N. E. Molau, J. Fockler, and S. Gregory, (1997),“BBU Design of Linear Induction Accelerator Cells forRadiography Application,” 1997 Particle AcceleratorConference, Vancouver, BC, Canada, May 12–16.

Sheaffer, M. K., S. C. Keeton, M. E. Wangler, L. E. Fischer,and Y. Y. Liu (1997), “Recent Developments in FissileMaterial Exemptions for Shipping Packages,”Proceedings of the American Nuclear Society TopicalMeeting on Criticality Safety Challenges in the NextDecade, September.

Simon, J., S. Saffer, and C.-J. Kim (1997), “A Liquid-filledMicrorelay with a Moving Mercury Microdrop,”Journal of Microelectromechanical Systems,September, Vol. 6, No. 3, pp. 208–216.

Simon, J., L.-S. Huang, B. Sridharan, and C.-J. Kim(1997), “Microgasketing and Room TemperatureWafer Joining for Liquid-Filled MEMS Devices,”Proceedings of the American Society of MechanicalEngineers International Mechanical EngineeringCongress and Exposition, Dallas, Texas, November,DSC-Vol. 62/HTD-Vol. 354, pp. 29–34.

Souza, P., and D. Chinn (1998), “Ultrasonic Measurementof KDP Solution Concentration,” ASNT 1998 FallConference, Nashville, Tennessee, October 19–23.

Spellman, G. P., R. Jayakumar, and R. P. Reed (1998),“Thermo-Mechanical Properties of ITER Buffer ZoneCandidate Materials,” Cryogenics, January, Vol. 38,No. 1, p. 33.

Stolz C. J., J. M. Yoshiyama, Z. L. Wu, A. Salleo, J. Green,and R. Krupka (1998), “Characterization of Nodularand Thermal Defects in Hafnia/Silica MultilayerCoatings Using Optical, Photothermal, and AtomicForce Microscopy,” Laser-Induced Damage in OpticalMaterials: 1997, G. J. Exarhos, A. H. Guenther, M. R.Kozlowski, and M. J. Soileau, eds., SPIE Proceedings,Vol. 3244, pp. 475–483.

Sugihwo, F. M., C. Larson, and J. S. Harris, Jr. (1998),“Micromachined Widely Tunable Vertical Cavity LaserDiodes,” Journal of Microelectromechanical Systems,March, Vol. 7, No. 1, pp. 48–55.

Sugihwo, F., M. C. Larson, and J. S. Harris, Jr. (1997),“Whispering Gallery Mode Operation in AirgapVertical Cavity Laser Structure,” Electronics Letters,August, Vol. 33, No. 17, pp. 1467–1468.

Sugihwo, F., M. C. Larson, and J. S. Harris, Jr. (1998),“Simultaneous Optimization of MembraneReflectance and Tuning Voltage for Tunable VerticalCavity Lasers,” Applied Physics Letters, January, Vol.72, No. 1, pp. 10–12.

Swan, J., D. Behne, C. M. Kendall, R. Yamamoto, T.Yokota, and J. Tanabe (1997), “Design andPerformance of the PEP-II B-Factory HER QD4Quadrupole Magnet,” Proceedings of the FifteenthInternational Conference on Magnet Technology,Science Press, China, October.

Syn, C. K., D. R. Lesuer, and O. D. Sherby (1997),“Thermo-Mechanical Processing and Properties of aDuctile Iron,” Thermo-Mechanical Processing andMechanical Properties of Hypereutectoid Steels andCast Irons, D. R. Lesuer, C. K. Syn and O. D. Sherby,eds., TMS, Warrendale, Pennsylvania, pp. 117–125.

Syn, C. K., D. R. Lesuer, T. G. Nieh, H. S. Yang, K. R.Brown, R. O. Kaibyshev, and E. N. Petrov (1998),“Roll Forming Technology for ManufacturingAxisymetric Automotive Components,” AutomotiveAlloys II, S. Das, ed., TMS, Warrendale, Pennsylvania,pp. 173–183.

Taleff, E. M., B. L. Bramfitt, C. K. Syn, D. R. Lesuer, and O.D. Sherby (1997), “Mechanical Behavior of anUltrahigh-Carbon Steel Exhibiting a Damask Pattern,”Thermo-Mechanical Processing and MechanicalProperties of Hypereutectoid Steels and Cast Irons,D. R. Lesuer, C. K. Syn and O. D. Sherby, eds., TMS,Warrendale, Pennsylvania, pp. 189–198.

Taleff, E. M., C. K. Syn, D. R. Lesuer, and O. D. Sherby(1997), “A Comparison of Mechanical Behavior inPearlitic and Spheroidized Hypereutectoid Steels,”Thermo-Mechanical Processing and MechanicalProperties of Hypereutectoid Steels and Cast Irons,”D. R. Lesuer, C. K. Syn and O. D. Sherby, eds., TMS,Warrendale, Pennsylvania, pp. 127–142.

Thomas, G. (1998), “Ultrasonic Evaluation of Flood GateTendons,” International Syposium on NDE Techniquesfor Aging Infrastructure and Manufacturing, SanAntonio, Texas, January.

Thomas, N., J. Wolfe, and J. Farmer (1998), “ProtectedSilver Coating for Astronomical Mirrors,” SPIEInternational Symposium on Astronomical Telescopeand Instrumentation Symposium, Kona, Hawaii, Vol.3553, March.

Tietbohl, G. L., P. M. Bell, R. M. Hamilton, J. B. Horner, R.L. Horton, A. P. Ludwigsen, J. L. Miller, W. H. Olson,C. S. Patel, D. M. Pennington, M. D. Vergino, and T. L.Weiland (1998), “Engineering the Petawatt Laser intoNova,” SPIE Photonics West Conference, San Jose,California, January 24–30.



Tripathi, B. P., and Q. A. Hossain (1997), “A CriticalReview of Methods to Evaluate Response ofStructures Subjected to Aircraft Impact,” Proceedingsof the Fourteenth Conference on StructuralMechanics in Reactor Technology, August 27, Vol. 6,Division J.

Watkins, B. E., R. S. Upadhye, and C. O. Pruneda (1997),“Recent Advances in the Molten Salt Technology forthe Destruction of Energetic Materials,” Proceedingsof the Air and Waste Management AssociationNinetieth Annual Meeting and Exposition, Toronto,Canada, June 8–13.

Watson, J. (1997), “Tip Tilt Correction System forAstronomical Telescopes Using Adaptive Control,”1997 IEEE WESCON/97 Conference, November,Record Catalog Number 97CH36148.

Westerberg, K. W., T. C. Meier, M. A. McClelland, D. G.Braun, L. V. Berzins, T. M. Anklam, and J. Storer(1997), “Analysis of the E-Beam Evaporation ofTitanium and Ti-6Al-4V,” Electron Beam Melting andRefining State of the Art 1997, R. Bakish, ed., BakishMaterials Corp., Englewood, New Jersey, pp.208–221.

Witte, M. C., J. Hovingh, G. C. Mok, S. S. Murty, T. F. Chen,and L. E. Fischer (1997), “Summary and Evaluationof Low-Velocity Impact Tests of Solid Steel Billet OntoConcrete Pads,” NUREG/CR-6608, U.S. NuclearRegulatory Commission.

Woods, B., M. Yan, J. DeYoreo, M. Kozlowski, H. Radousky,and Z. L. Wu (1998), “Photothermal Mapping ofDefects in the Study of Bulk Damage in KDP,” Laser-Induced Damage in Optical Materials: 1997, G. J.Exarhos, A. H. Guenther, M. R. Kozlowski, and M. J.Soileau, eds., SPIE Proceedings, Vol. 3244, pp.242–248.

Wu, Z. L. (1998), “Photothermal Sensing Techniques forThin Film Characterization,” in MaterialsCharacterization and Optical Probe Techniques,Critical Review of Optical Science and Technology, CR69, pp. 326–356.

Wu, Z. L., J. Green, T. Yang, and R. Krupka (1997), “Non-Destructive Evaluation of Large Aperture Optics WithUltra-Low Optical Absorption,” ThirteenthSymposium on Thermophysical Properties, Boulder,Colorado, June 22–27.

Young, D. J., V. Malba, J. J. Ou, B. E. Boser, and A. F.Bernhardt (1997), “Monolithic High-PerformanceThree-Dimensional Coil Inductors For WirelessCommunication Applications,” Proceedings of theIEEE International Electron Devices Meeting,Washington, DC, December 7–10.

Zhao, Q., Z. L. Wu, M. Thomsen, and Y. Han (1998),“Interfacial Effects on the Transient Temperature Riseof Multilayer Coatings Induced by a Short-PulseLaser Irradiation,” Laser-Induced Damage in OpticalMaterials: 1997, G. J. Exarhos, A. H. Guenther, M. R.Kozlowski, and M. J. Soileau, eds., SPIE Proceedings,Vol. 3244, pp. 491–498.

Zocher, M. A., D. H. Allen, and S. E. Groves (1997), “AThree-Dimensional Finite Element Formulation forThermoviscoelastic Orthotropic Media,” InternationalJournal for Numerical Methods in Engineering, June,Vol. 40.

Zocher, M. A., D. H. Allen, and S. E. Groves (1997),“Stress Analysis of a Matrix-Cracked ViscoelasticLaminate,” International Journal of Solids andStructures, Vol. 34 No. 25.

FY 98 PB-9

Author Index


Author Index

FY98 AI-1

Ackler, H........................................................2-37Afeyan, B. B. ....................................................4-1Allen, K. A........................................................2-1An, J. R. .........................................................6-27Ashby, E.........................................................4-13Aufderheide, M. B. .........................................5-61Avalle, C. A. ...................................................6-23Azevedo, S. ....................................................5-39

Balooch, M.....................................................2-23Bettencourt, K. A. ............................................2-1Blaedel, K. L. ...................................................3-9Bland, M. F.....................................................4-29Brase, J. M. ...................................................5-39Brown, A. A......................................................6-1Brugger, S. T. ...................................................4-7Bryan, Jr., S. R...............................................6-17Burke, G. .........................................................4-7Butler, J. A.......................................................2-1

Campbell, E. W. .............................................6-27Candy, J. V. ...............................................1-3, 5-1Caporaso, G. J. ..............................................4-85Casey, J. ..........................................................6-1Castor, J. I. ....................................................6-23Chen, Y.-J.......................................................4-85Chien, C. ........................................................2-11Chinn, D. ..............................2-23, 5-1, 5-19, 5-23Ciarlo, D. R....................................2-5, 2-23, 4-13Clague, D. S. ..................................................2-19Clark, G. A. ......................................................1-3Cooper, G. A...................................................2-31

Darrow, C. B. .................................................2-45Davis, P. J. .......................................................3-9De Groot, A. J. ........................................1-1, 4-47DenBaars, S...................................................2-31Deng, C. ........................................................2-11Deri, R. J. ...............................................1-1, 2-49Dunlap, J. E. ..................................................4-71Dupuy, P. C. .....................................................3-9

Erlandson, A. C..............................................4-37Elgorria, I. .....................................................2-11

Feit, M. D.........................................................4-1Ferencz, R. M.................................................4-47Fugina, J. M...................................................5-27

Garcia, M. ......................................................4-33Gilman, A.......................................................2-45Goodman, D. M. .............................................5-61Graff, R. T. .....................................................2-27Gresho, P. M. .................................................4-81Groves, S. E. .........................................2-23, 5-23

Hagans, K. G. .................................................2-17Haigh, R. E. ..........................................2-13, 2-17Haskins, J. J. ........................................5-23, 5-27Haupt, D. L. ...................................................5-23Hayes, J. P. ....................................................2-27Hernandez, M. A. .............................................4-7Hoehler, M. ......................................................1-3Hoover, C. G...................................................4-47Huber, R. D. .....................................................5-1

Jackson, J. A..................................................5-61Jancaitis, K. S. ...............................................4-37Jankowski, A. F. .............................................2-27Jensen, S. A. ....................................................3-1Johansson, E. M.............................................5-61Johnson, R. R.................................................1-11Jones, L. M. .....................................................2-7

Kallman, J. S. ..................................4-1, 4-7, 4-13Kay, G. J...........................................................6-7Kinney, J. .......................................................5-23Krulewich, D. A................................................3-3

LaChappel, M. J. ............................................2-23Lane, S. M......................................................2-45Larson, M.C. ..................................................2-49Lavietes, A. D. ...............................................5-27Law, B. P..........................................................2-5LeBlanc, M.......................................................6-7Lee, A. P. .......................................2-7, 2-11, 2-45Lee, Y. L.........................................................4-47Lehew, S. L. ...................................................2-31Le Sage, G. P..................................................4-19Lesuer, D. R.....................................................6-7LeTouze, G. ....................................................4-37Lin, J. I. .........................................................4-55Logan, C. M. .........................................5-23, 5-27Lowry, M. E. .........................................2-13, 2-49

Marshall, C. D................................................4-37Mast, J. E. ............................................4-23, 5-39Mayhall, D. J..................................................4-29McCallen, D. B. ................................................1-3McConaghy, C. F. ..................2-7, 2-11, 2-13, 2-45Meyer, G. A. ...................................................2-31Miles, R. R. ......................................................2-1Mishra, U. ......................................................2-31Molau, N. E. ...................................................4-77Morales, K. E. ................................................5-27Morse, J. D. ..........................................2-23, 2-27

Nelson, S. D..........................................4-23, 4-71Nikkel, D. J., Jr ................................................6-1Nukala, P. K. V. V............................................4-65

Author Index

Perry, R. L. ....................................................5-13Piscotty, M. A...................................................3-9Pocha, M. D. .........................................2-41, 2-49Poole, B. R.....................................................4-85Puso, M. A. ....................................................4-59

Ratowsky, R. P..................................................4-1Richardson, R. A. ...........................................4-19Roberts, R. S...........................................5-9, 5-13Robey, H. F.....................................................5-19Rosenbury, E. T. .............................................5-39Rotter, M. D. ..................................................4-37

Sanders, D. L. ................................................6-17Sargis, P. D. ...................................................4-77Satcher, Jr., J. H. ...........................................2-45Schmid, G. J...................................................5-27Schneberk, D. J. ............................................5-27Schumann, D. L. ..............................................2-1Sengupta, S. K. ..............................................5-51Seznec, S. ......................................................4-37Shapiro, A. B..................................................4-81Sherwood, R. J...............................................4-47Sigmon, T. W. .................................................2-31Simon, J...........................................................2-7Smart, J. C.....................................................1-15

Soltani, P. ......................................................5-27Souza, P. R.....................................................5-19Speck, D. E. ...................................................4-47Spicer, J...........................................................5-1Springer, K.....................................................5-27Steich, D. .........................................................4-7Stuart, B. C. .....................................................3-1Swartz, K. ......................................................5-27Sweider, D. ....................................................4-77Swierkowski, S. P...........................................2-37

Thomas, G. H. .........................................4-13, 5-1Toet, D. ..........................................................2-31Trevino, J. C. ....................................................2-7

Updike, E. O...................................................5-27

Wang, A. W.....................................................2-45Wang, L. ...............................................4-23, 4-85Warrick, A. L..................................................4-23Waters, A. ......................................................5-23White, D. A. .....................................................4-7

Zapata, L. E. ..................................................4-37Zywicz, E. ......................................................4-47

Engineering Research Development and TechnologyAI-2

Technical Information DepartmentLawrence Livermore National LaboratoryUniversity of CaliforniaLivermore, California 94551

Technical Information DepartmentLawrence Livermore National LaboratoryUniversity of CaliforniaLivermore, California 94551

Eng

ineerin

g R

esearch, D

evelop

men

t and

Techn

olo

gy • FY

98 • UC

RL 53868-98

Cover_qk 9/1/99 3:56 PM Page 1